My Climate Change News

The global methane (CH₄) budget is becoming an increasingly important component for managing realistic pathways to mitigate climate change. This relevance, due to a shorter atmospheric lifetime and a stronger warming potential than carbon dioxide, is challenged by the still unexplained changes of atmospheric CH₄ over the past decade. Emissions and concentrations of CH₄ are continuing to increase, making CH₄ the second most important human-induced greenhouse gas after carbon dioxide. Two major difficulties in reducing uncertainties come from the large variety of diffusive CH₄ sources that overlap geographically, and from the destruction of CH₄ by the very short-lived hydroxyl radical (OH). To address these difficulties, we have established a consortium of multi-disciplinary scientists under the umbrella of the Global Carbon Project to synthesize and stimulate research on the methane cycle, and producing regular (∼ biennial) updates of the global methane budget. This consortium includes atmospheric physicists and chemists, biogeochemists of surface and marine emissions, and socio-economists who study anthropogenic emissions. Following Kirschke et al. (2013), we propose here the first version of a living review paper that integrates results of top-down studies (exploiting atmospheric observations within an atmospheric inverse-modelling framework) and bottom-up models, inventories and data-driven approaches (including process-based models for estimating land surface emissions and atmospheric chemistry, and inventories for anthropogenic emissions, data-driven extrapolations).

For the 2003–2012 decade, global methane emissions are estimated by top-down inversions at 558 Tg CH₄ yr^-1, range 540–568. About 60 % of global emissions are anthropogenic (range 50–65 %). Since 2010, the bottom-up global emission inventories have been closer to methane emissions in the most carbon-intensive Representative Concentrations Pathway (RCP8.5) and higher than all other RCP scenarios. Bottom-up approaches suggest larger global emissions (736 Tg CH₄ yr^-1, range 596–884) mostly because of larger natural emissions from individual sources such as inland waters, natural wetlands and geological sources. Considering the atmospheric constraints on the top-down budget, it is likely that some of the individual emissions reported by the bottom-up approaches are overestimated, leading to too large global emissions. Latitudinal data from top-down emissions indicate a predominance of tropical emissions (∼ 64 % of the global budget, < 30◦ N) as compared to mid (∼ 32 %, 30–60◦ N) and high northern latitudes (∼ 4 %, 60–90◦ N). Top-down inversions consistently infer lower emissions in China (∼ 58 Tg CH₄ yr^-1, range 51–72, −14 %) and higher emissions in Africa (86 Tg CH₄ yr^-1, range 73–108, >+19 %) than bottom-up values used as prior estimates. Overall, uncertainties for anthropogenic emissions appear smaller than those from natural sources, and the uncertainties on source categories appear larger for top-down inversions than for bottom-up inventories and models.

The most important source of uncertainty on the methane budget is attributable to emissions from wetland and other inland waters. We show that the wetland extent could contribute 30–40 % on the estimated range for wetland emissions. Other priorities for improving the methane budget include the following: (i) the development of process-based models for inland-water emissions, (ii) the intensification of methane observations at local scale (flux measurements) to constrain bottom-up land surface models, and at regional scale (surface networks and satellites) to constrain top-down inversions, (iii) improvements in the estimation of atmospheric loss by OH, and (iv) improvements of the transport models integrated in top-down inversions. The data presented here can be downloaded from the Carbon Dioxide Information Analysis Center (http://doi.org/10.3334/CDIAC/GLOBAL_ METHANE_BUDGET_2016_V1.1) and the Global Carbon Project.

The works published in this journal are distributed under the Creative Commons Attribution 3.0 License. This license does not affect the Crown copyright work, which is re-usable under the Open Government Licence (OGL). The Creative Commons Attribution 3.0 License and the OGL are interoperable and do not conflict with, reduce or limit each other.

The surface dry air mole fraction of atmospheric methane (CH₄) reached 1810 ppb in 2012 (Fig. 1). This level, 2.5 times larger than in 1750, results from human activities related to agriculture (livestock, rice cultivation), fossil fuel usage and waste sectors, and from climate and CO₂-changes affecting natural emissions (Ciais et al., 2013). Atmospheric CH₄ is the second most impactful anthropogenic greenhouse gas after carbon dioxide (CO₂) in terms of radiative forcing. Although its global emissions, estimated at around 550 Tg CH₄ yr^-1 (Kirschke et al., 2013), are only 4 % of the global CO₂ anthropogenic emissions in units of carbon mass flux, atmospheric CH₄ has contributed 20 % (∼ 0.48 W m^-2,) of the additional radiative forcing accumulated in the lower atmosphere since 1750 (Ciais et al., 2013).

This is because of the larger warming potential of methane compared to CO₂, about 28 times on a 100-year horizon as re-evaluated by the Intergovernmental Panel on Climate Change (IPCC) 5th Assessment Report (AR5) (when using Global Warming Potential metric; Myhre et al., 2013). Changes in other chemical compounds (such as NOxor CO) also influence the forcing of methane through changes in its lifetime. From an emission point of view, the radiative impact attributed to CH₄ emissions is about 0.97 W m.

This is because emission of CH₄ leads to production of ozone, of stratospheric water vapour, and of CO₂, and importantly affects its own lifetime (Myhre et al., 2013; Shindell et al., 2012). CH₄ has a short lifetime in the atmosphere (∼9 years for the modern inventory; Prather et al., 2012), and a stabilization or reduction of CH₄ emissions leads rapidly to a stabilization or reduction of methane radiative forcing. Reduction in CH₄ emissions is therefore an effective option for climate change mitigation. Moreover, CH₄ is both a greenhouse gas and an air pollutant, and as such covered by two international conventions: the United Nations Framework Convention on Climate Change (UNFCCC) and the Convention on Long Range Transport of Air Pollution (CTRTAP).

Changes in the magnitude and timing (annual to interannual) of individual methane sources and sinks over the past decades are uncertain (Kirschke et al., 2013) with relative uncertainties (hereafter reported as min–max ranges) of 20– 30 % for inventories of anthropogenic emissions in each sector (agriculture, waste, fossil fuels) and for biomass burning, 50 % for natural wetland emissions and reaching 100 % or more for other natural sources (e.g. inland waters, geological). The uncertainty in the global methane chemical loss by OH, the predominant sink, is estimated between 10 % (Prather et al., 2012) and 20 % (Kirschke et al., 2013), implying a similar uncertainty in global methane emissions as other sinks are much smaller and the atmospheric growth rate is well defined (Dlugokencky et al., 2009). Globally, the contribution of natural emissions to the total emissions is reasonably well quantified by combining lifetime estimates with reconstructed preindustrial atmospheric methane concentrations from ice cores (e.g. Ehhalt et al., 2001). Uncertainties in emissions reach 40–60 % at regional scale (e.g. for South America, Africa, China and India). Beyond the intrinsic value of characterizing the biogeochemical cycle of methane, understanding the evolution of the methane budget has strong implications for future climate emission scenarios. Worryingly, the current emission trajectory is tracking the warmest of all IPCC scenarios, the RCP8.5, and is clearly inconsistent with lower temperature scenarios, which show substantial to large reductions of methane emissions (Collins et al., 2013).

Reducing uncertainties in individual methane sources, and thus in the overall methane budget, is not an easy task for, at least, four reasons. First, methane is emitted by a variety of processes that need to be understood and quantified separately, both natural or anthropogenic, point or diffuse sources, and associated with three main emission processes (biogenic, thermogenic and pyrogenic). Among them, several important anthropogenic CH₄ emission sources are poorly reported. These multiple sources and processes require the integration of data from diverse scientific communities to assess the global budget. Second, atmospheric methane is removed by chemical reactions in the atmosphere involving radicals (mainly OH), which have very short lifetimes (typically 1 s). Although OH can be measured locally, its spatiotemporal distribution remains uncertain at regional to global scales, which cannot be assessed by direct measurements. Third, only the net methane budget (sources – sinks) is constrained by the precise observations of the atmospheric growth rate (Dlugokencky et al., 2009), leaving the sum of sources and the sum of sinks uncertain. One simplification for CH₄ compared to CO₂ is that the oceanic contribution to the global methane budget is very small (∼ 1–3 %), making source estimation mostly a continental problem (USEPA, 2010a). Finally, we lack observations to constrain (1) process models that produce estimates of wetland extent (Stocker et al., 2014; Kleinen et al., 2012) and emissions (Melton et al., 2013; Wania et al., 2013), (2) other inland water sources (Bastviken et al., 2011), (3) inventories of anthropogenic emissions (USEPA, 2012; EDGARv4.2FT2010, 2013), and (4) atmospheric inversions, which aim at representing or estimating the different methane emissions from global to regional scales (Houweling et al., 2014; Kirschke et al., 2013; Bohn et al., 2015; Spahni et al., 2011; Tian et al., 2016). Finally, information contained in the ice core methane records has only been used in a few studies to evaluate process models (Zürcher et al., 2013; Singarayer et al., 2011).

The regional constraints brought by atmospheric sampling on atmospheric inversions are significant for northern mid latitudes thanks to a number of high-precision and highaccuracy surface stations (Dlugokencky et al., 2011). The atmospheric observation density has improved in the tropics with satellite-based column-averaged methane mixing ratios (Buchwitz et al., 2005b; Frankenberg et al., 2005; Butz et al., 2011). However, the optimal usage of satellite data remains limited by systematic errors in satellite retrievals (Bergamaschi et al., 2009; Locatelli et al., 2015). The development of low-bias observations system from space, such as active lidar technics, is promising to overcome these issues (Kiemle et al., 2014). The partition of regional emissions by processes remains very uncertain today, waiting for the development or consolidation of measurements of more specific tracers, such as methane isotopes or ethane, dedicated to constrain the different methane sources or groups of sources (e.g. Simpson et al., 2012; Schaefer et al., 2016; Hausmann et al., 2016).

The Global Carbon Project (GCP) aims at developing a complete picture of the carbon cycle by establishing a common, consistent scientific knowledge to support policy debate and actions to mitigate the rate of increase of greenhouse gases in the atmosphere (http://www.globalcarbonproject. org). The objective of this paper is to provide an analysis and synthesis of the current knowledge about the global and regional methane budgets by gathering results of observations and models and by extracting from these the robust features and the uncertainties remaining to be addressed. We combine results from a large ensemble of bottom-up approaches (process-based models for natural wetlands, data-driven approaches for other natural sources, inventories of anthropogenic emissions and biomass burning, and atmospheric chemistry models) and of top-down approaches (methane atmospheric observing networks, atmospheric inversions inferring emissions and sinks from atmospheric observations and models of atmospheric transport and chemistry). The focus here is on decadal budgets, leaving in-depth analysis of trends and year-to-year changes to future publications. This paper is built on the principle of a living review to be published at regular intervals (e.g. every two years) and will synthesize and update new annual data, the introduction of new data products, model development improvements, and new modelling approaches to estimate individual components contributing to the CH₄ budget.

The work of Kirschke et al. (2013) was the first GCP-like CH₄ budget synthesis. Kirschke et al. (2013) reported decadal mean CH₄ emissions and sinks from 1980 to 2009 based on bottom-up and top-down approaches. Our new analysis, and our approach for the living review budget, will report methane emissions for three targeted time periods:

the last calendar decade (2000–2009, for this paper),

the last available decade (2003–2012, for this paper), and

the last available year (2012, for this paper). Future efforts will also focus on retrieving budget data as recent as possible.

Five sections follow this introduction. Section 2 presents the methodology to treat and analyse the data streams. Section 3 presents the current knowledge about methane sources and sinks based on the ensemble of bottom-up approaches reported here (models, inventories, data-driven approaches). Section 4 reports the atmospheric observations and the top-down inversions gathered for this paper. Section 5, based on Sects. 3 and 4, provides an analysis of the global methane budget (Sect. 5.1) and of the regional methane budget (Sect. 5.2). Finally Sect. 6 discusses future developments, missing components and the largest remaining uncertainties after this update on the global methane budget.

Unless specified, the methane budget is presented in teragrammes of CH₄ per year (1 Tg CH₄ yr^-1 >= 1012 g CH₄ yr^-1), methane concentrations as dry air mole fractions in parts per billion (ppb) and the methane annual increase, GATM, in ppb yr^-1. In the different tables, we present mean values and ranges for the last calendar decade (2000–2009, for this paper), the period 2003–2012, together with results for the last available year (2012, for this paper). Results obtained from the previous synthesis are also given (Kirschke et al., 2013, for this paper). Following Kirschke et al. (2013) and considering the relatively small and variable number of studies generally available for individual numbers, uncertainties are reported as minimum and maximum values of the gathered studies in brackets. In doing so, we acknowledge that we do not take into account all the uncertainty of the individual estimates (when provided). This means that the full uncertainty range may be greater than the range provided here. These minimum and maximum values are those calculated using the boxplot analysis presented below and thus excluding identified outliers when existing.

The CH₄ emission estimates reported in this paper, derived mainly from statistical calculations, are given with up to three digits for consistency across all budget flux components and to ensure conservation of quantities when aggregated into flux categories in Table 2 (and regional sources in Table 4). However, the reader should keep in mind the associated uncertainties and acknowledge a two-digit global methane budget.

Common data analysis procedures have been applied to the different bottom-up models, inventories and atmospheric inversions whenever gridded products exist. The monthly or yearly fluxes (emissions and sinks) provided by different groups were processed similarly. They were re-gridded on a common grid (1◦ × ¹◦) and converted into the same units (Tg CH₄ per grid cell). For coastal pixels of land fluxes, to avoid allocating land emissions into oceanic areas when regridding the model output, all emissions were re-allocated to the neighbouring land pixel. The opposite was done for ocean fluxes. Monthly, annual and decadal means were computed from the gridded 1◦ by 1◦ maps.

Most budgets are presented as boxplots, which have been created using routines in IDL language, provided with the standard version of the IDL software. The values presented in the following are calculated using the classical conventions of boxplots including quartiles (25 %, median, 75 %), outliers, and minimum and maximum values (without the outliers). Outliers are determined as values below the first quartile minus 3 times the interquartile range or values above third quartile plus 3 times the interquartile range. Identified outliers (when existing) are plotted as stars on the different figures proposed. The mean values are reported in the tablesand represented as “>+” symbols in the figures.

Geographically, emissions are reported for the global scale, for three latitudinal bands (< 30, 30–60, 60–90◦ N, only for gridded products) and for 15 regions (oceans and 14 continental regions, see Sect. 5 and Fig. 7 for region map). As anthropogenic emissions are reported at country level, we chose to define the regions based on a country list (Supplement Table S1). This approach is compatible with all topdown and bottom-up approaches providing gridded products as well. The number of regions was chosen to be close to the widely used TransCom intercomparison map (Gurney et al., 2004), but with subdivisions to isolate important countries for the methane budget (China, India, USA and Russia). Therefore, the new region map defined here is different from the TransCom map but more adapted to the methane cycle. One caveat is that the regional totals are not directly comparable with other studies reporting methane emissions on the TransCom map (as in Kirschke et al., 2013, for example), although the names of some regions are the same.

Bottom-up estimates of methane emissions rely on models for individual processes (e.g. wetlands) or on inventories representing different source types (e.g. gas emissions). Chemistry transport models generally represent methane sinks individually in their chemical schemes (Williams et al., 2012). Therefore, it is possible to represent the bottom-up global methane budget for all individual sources. However, by construction, the total methane emissions derived from a combination of independent bottom-up estimates are not constrained.

For atmospheric inversions (top-down), the situation is different. Atmospheric observations provide a constraint on the global source, given a fairly strong constraint on the global sink derived using a proxy tracer such as methyl chloroform (Montzka et al., 2011). The inversions reported in this work solve either for a total methane flux (e.g. Pison et al., 2013) or for a limited number of flux categories (e.g. Bergamaschi et al., 2013). Indeed, the assimilation of CH₄ observations alone, as reported in this synthesis, cannot fully separate individual sources, although sources with different locations or temporal variations could be resolved by the assimilated atmospheric observations. Therefore, following Kirschke et al. (2013), we have defined five broad categories for which top-down estimates of emissions are given: natural wetlands, agriculture and waste emissions, fossil fuel emissions, biomass and biofuel burning emissions, and other natural emissions (other inland waters, wild animals, wildfires, termites, land geological sources, oceanic sources (geological and biogenic), and terrestrial permafrost). Global and regional methane emissions per source category were obtained directly from the gridded optimized fluxes if an inversion solved for the GCP categories. Alternatively, if the inversion solved for total emissions (or for different categories embedding GCP categories), then the prior contribution of each source category at the spatial resolution of the inversion was scaled by the ratio of the total (or embedding category) optimized flux divided by the total (or embedding category) prior flux (Kirschke et al., 2013). Also, the soil uptake was provided separately in order to report the total surface emissions and not net emissions (sources minus soil uptake). For bottom-up, some individual sources can be found gridded in the literature (anthropogenic emissions, natural wetlands), but some others are not gridded yet (e.g. inland waters, geological, oceanic sources). The regional bottom-up methane budget per source category is therefore presented only for gridded categories (all but the “other natural” category).

In summary, bottom-up models and inventories are presented for all individual sources and for the five broad categories defined above at global scale, and only for four broad categories at regional scale. Top-down inversions are reported globally and regionally for the five broad categories of emissions.

Here we provide a complete review of all methane sources and sinks based on an ensemble of bottom-up approaches from multiple sources: process-based models, inventories, and data-driven methods. For each source, a description of the involved emitting process(es) is given, together with a brief description of the original datasets (measurements, models) and the related methodology. Then, the estimate for the global source and its range is given and analysed. Detailed descriptions of the datasets can be found elsewhere (see references of each component in the different subsections and tables).

Methane is emitted by a variety of sources in the atmosphere. These can be sorted by emitting process (thermogenic, biogenic or pyrogenic) or by anthropogenic vs. natural origin. Biogenic methane is the final product of the decomposition of organic matter by Archaea in anaerobic environments, such as water-saturated soils, swamps, rice paddies, marine sediments, landfills, waste-water facilities, or inside animal intestines. Thermogenic methane is formed on geological timescales by the breakdown of buried organic matter due to heat and pressure deep in the Earth’s crust. Thermogenic methane reaches the atmosphere through marine and land geologic gas seeps and during the exploitation and distribution of fossil fuels (coal mining, natural gas production, gas transmission and distribution, oil production and refinery). Finally, pyrogenic methane is produced by the incomplete combustion of biomass. Peat fires, biomass burning in deforested or degraded areas, and biofuel usage are the largest sources of pyrogenic methane. Methane hydrates, icelike cages of methane trapped in continental shelves and below sub-sea and land permafrost, can be of biogenic or thermogenic origin. Each of the three process categories has both anthropogenic and natural components. In the following, we choose to present the different methane sources depending on their anthropogenic or natural origin, which seems more relevant for planning climate mitigation activities. However this choice does not correspond exactly to the definition of anthropogenic and natural used by UNFCCC and IPCC guidelines, where, for pragmatic reasons, all emissions from managed land are reported as anthropogenic, which is not the case here. For instance, we consider all wetlands in the natural emissions whereas there are managed wetlands.

Various human activities lead to the emissions of methane to the atmosphere. Agricultural processes under anaerobic conditions such as wetland rice cultivation and livestock (enteric fermentation in animals, and the decomposition of animal wastes) emit biogenic CH₄, as does the decomposition of municipal solid wastes. Methane is also emitted during the production and distribution of natural gas and petroleum and is released as a byproduct of coal mining and incomplete fossil fuel and biomass combustion (USEPA, 2016).

Emission inventories were developed to generate bottom-up estimates of sector-specific emissions by compiling data on human activity levels and combining them with the associated emission factors.

An ensemble of individual inventories was gathered here to estimate anthropogenic methane emissions. We also refer to the extensive assessment report of the Arctic Monitoring and Assessment Programme (AMAP) published in 2015 on “Methane as Arctic climate forcer” (Höglund-Isaksson et al., 2015), which provides a detailed presentation and description of methane inventories and global scale estimates for the year 2005 (see their chap. 5 and in particular their Tables 5.1 to 5.5).

The main three bottom-up global inventories covering all anthropogenic emissions are from the United States Environmental Protection Agency, USEPA (2012, 2006), the Greenhouse gas and Air pollutant Interactions and Synergies (GAINS) model developed by the International Institute for Applied Systems Analysis (IIASA) (Höglund-Isaksson, 2012) and the Emissions Database for Global Atmospheric Research (EDGARv4.1, 2010; EDGARv4.2FT2010, 2013).

The latter is an inventory compiled by the European Commission Joint Research Centre (EC-JRC) and Netherland’s Environmental Assessment Agency (PBL). These inventories report the major sources of anthropogenic methane emissions: fossil fuel production, transmission and distribution; livestock (enteric fermentation and manure management); rice cultivation; solid waste and waste water. However, the level of detail provided by country and by sector varies between inventories, as these inventories do not consider the same number of geographical regions and source sectors (Höglund-Isaksson et al., 2015, see their Table 5.2). In these inventories, methane emissions for a given region/country and a given sector are usually calculated as the product of an activity level, an emission factor for this activity and an abatement coefficient to account for regulations implemented to control emissions if existing (see Eq. 5.1 of Höglund-Isaksson et al., 2015; IPCC, 2006). The integrated emission models USEPA and GAINS provide estimates every 5 or 10 years for both past and future periods, while EDGAR provides annual estimates only for past emissions. These datasets differ in their assumptions and the data used for the calculation; however, they are not completely independent as they follow the IPCC guidelines (IPCC, 2006). While the USEPA inventory adopts the emissions reported by the countries to the UNFCCC, EDGAR and the GAINS model produced their own estimates using a consistent approach for all countries. As a result, the latter two approaches need large country-specific information or, if not available, they adopt IPCC default factors or emission factors reported to UNFCCC (Olivier et al., 2012; Höglund-Isaksson, 2012). Here, we also integrate the Food and Agriculture Organization (FAO) dataset, which provides estimates of methane emissions at country level but only for agriculture (enteric fermentation, manure management, rice cultivation, energy usage, burning of crop residues and of savannahs) and land use (biomass burning) (FAO, 2016). It will hereafter be referred as FAO-CH₄. FAO-CH₄ uses activity data from the FAOSTAT database as reported by countries to National Agriculture Statistical Offices (Tubiello et al., 2013) and mostly the Tier 1 IPCC methodology for emission factors (IPCC, 2006), which depend on geographic location and development status of the country. For manure, the necessary country-scale temperature was obtained from the FAO global agro-ecological zone database (GAEZv3.0, 2012).

We use the following versions of these inventories: version EDGARv4.2FT2010 that provides yearly gridded emis sions by sectors from 2000 to 2010 (Olivier and Janssens-Maenhout, 2012; EDGARv4.2FT2010, 2013), version 5a of the GAINS model (Höglund-Isaksson, 2012) that assumes current legislation for air pollution for the future, the revised estimates of 2012 from the USEPA (2012), and finally the FAO emission database accessed in April 2016. Further details of the inventories used in this study are provided in Table 1. Overall, only EDGARv4.2FT2010 and GAINS provide gridded emission maps by sectors, and only EDGAR provides gridded maps on a yearly basis, which ex plains why this inventory is the most used in inverse mod elling. These inventories are not all regularly updated. For the purpose of this study, the estimates from USEPA and GAINS have been linearly interpolated to provide yearly values, as provided by the EDGAR inventory. We also use the EDGARv4.2FT2012 data, which is an update of the time series of the country total emissions until 2012 (Rogelj et al., 2014; EDGARv4.2FT2012, 2014). This update has been developed based on EDGARv4.2FT2010 and uses IEA energy balance statistics (IEA, 2013) and NIR/CRF of UNFCCC (2013), as described in part III of IEA’s CO2 book by Olivier and Janssens-Maenhout (2014).

For this study, engaged before the update of EDGARv4.2 up to 2012, we built our own update from 2008 up to 2012 using FAO emissions to quantify CH₄ emissions from enteric fermentation, manure management and rice cultivation (described above) and BP statistical review of fossil fuel production and consumption (http://www.bp.com/) to update CH₄ emissions from coal, oil and gas sectors. In this inventory, called EDGARv4.2EXT, methane emissions after 2008 are set up equal to the FAO emissions (or BP statistics) of year t times the ratio between the mean EDGAR CH₄ emissions (EEDGARv4.₂) over 2006–2008 and the mean value of FAO emissions (VFAOin the following equation) (or BP statistics) over 2006–2008. For each emission sector, the country-specific emissions (EEDGARv4.2ext) in year (t ) are estimated following Eq. (1):(See original document for equation)

Other sources than those aforementioned are kept constant at the 2008 level. This extrapolation approach is necessary and often performed by top-down inversions to define prior emissions, because, up to now, global inventories such as sector-specific emissions in EDGAR database have not been updated on a regular basis. EC-JRC released, however, their update up to 2012 (EDGARv4.2FT2012) containing country total emissions, which allows evaluation of our extrapolation approach. The extrapolated global totals of EDGARv4.2EXT are within 1 % of EDGARv4.2FT2012.

3.1.2

Based on the ensemble of inventories detailed above, anthr3.1.3opogenic emissions are ∼ 352 [340–360] Tg CH₄ yr^-1 for the decade 2003–2012 (Table 2, including biomass and biofuel burning). For the 2000–2009 period, anthropogenic emissions are estimated at ∼ 338 [329–342] Tg CH₄ yr^-1. This estimate is consistent, albeit larger and with a smaller uncertainty range than Kirschke et al. (2013) for the 2000–2009 decade (331 Tg CH₄ yr^-1 [304–368]). Such differences are due to the different sets of inventories gathered. The range of our estimate (∼ 5 %) is smaller then the range reported in the AMAP assessment report (∼ 20 %) both because the latter was reporting more versions of the different inventories and projections, and because it was for the particular year 2005 and not for a decade as here.

Figure 2 presents the global methane emissions of anthropogenic sources (excluding biomass and biofuel burning) estimated and projected by the different inventories between 2000 and 2020. The inventories consistently estimate that about 300 Tg of methane was released into the atmosphere in 2000 by anthropogenic activities. The main discrepancy between the inventories is observed in their trend after 2005 with the lowest emissions projected by USEPA and the largest emissions estimated by EDGARv4.2FT2012. The increase in CH₄ emissions is mainly determined from coal mining, whose activity increased considerably in China from 2002 to 2012 (see Sect. 3.1.3).

Despite relatively good agreement between the inventories on total emissions from year 2000 onwards, large differences can be found at the sector and country levels (IPCC, 2014). Some of these discrepancies are detailed in the following sections.

For the fifth IPCC Assessment Report, four representative concentration pathways (RCPs) were defined RCP8.5, RCP6, RCP4.5 and RCP2.6 (the latter is also referred to as RCP3PD, where “PD” stands for peak and decline). The numbers refer to the radiative forcing by the year 2100 in W m^-2,. These four independent pathways developed by four individual modelling groups start from the identical base year 2000 (Lamarque et al., 2010) and have been harmonized with historical emissions up to 2005. An interesting feature is the fact that global emission inventories track closer to methane emissions in the most carbon-intensive scenario (RCP8.5) and that all other RCP scenarios remain below the inventories. This suggests the tremendous challenge of climate mitigation that lies ahead, particularly if current trajectories need to change to be consistent with pathways leading to lower levels of global warming (Fig. 2).

Most of the methane anthropogenic emissions related to fossil fuels come from the exploitation, transportation, and usage of coal, oil and natural gas. This geological and fossil type of emission (see natural source section) is driven by human activity. Additional emissions reported in this category include small industrial contributions such as production of chemicals and metals, and fossil fuel fires. Spatial distribution of methane emissions from fossil fuel is presented in Fig. 3 based on the mean gridded maps provided by EDGARv4.2FT2010 and GAINS over the 2003–2012 decade.

Global emissions of methane from fossil fuels and other industries are estimated from three global inventories in the range of 114–133 Tg CH₄ yr−1 for the 2003–2012 decade with an average of 121 Tg CH₄ yr^-1 (Table 2), but with a large difference in the rate of change depending on inventories. It represents on average 34 % (range 32–39 %) of the total global anthropogenic emissions.

During mining, methane is emitted from ventilation shafts, where large volumes of air are pumped into the mine to keep methane at a rate below 0.5 % to avoid accidental inflammation. To prevent the diffusion of methane in the mining working atmosphere, boreholes are made in order to evacuate methane. In countries of the Organization for Economic Cooperation and Development (OECD), methane recuperated from ventilation shafts is used as fuel, but in many countries it is still emitted into the atmosphere or flared, despite efforts for coal-mine recovery under the UNFCCC Clean Development Mechanisms (http://cdm.unfccc.int). Methane emissions also occur during post-mining handling, processing, and transportation. Some CH₄ is released from coal waste piles and abandoned mines. Emissions from these sources are believed to be low because much of the CH₄ would likely be emitted within the mine (IPCC, 2000).

Almost 40 % (IEA, 2012) of the world’s electricity is produced from coal. This contribution grew in the 2000s at the rate of several per cent per year, driven by Asian production where large reserves exist, but has stalled from 2011 to 2012. In 2012, the top 10 largest coal producing nations accounted for 88 % of total world emissions for coal mining. Among them, the top three producers (China, USA and India) produced two-thirds of the total (CIA, 2016).

Global estimates of methane emissions from coal mining show a large variation, in part due to the lack of comprehensive data from all major producing countries. The range of coal mining emissions is estimated at 18–46 Tg of methane for the year 2005, the highest value being from EDGARv4.2FT2010 and the lower from USEPA.

As announced in Sect. 3.1.2, coal mining is the main source explaining the differences observed between inventories at global scale (Fig. 2). Indeed, such differences are explained mainly by the different CH₄ emission factors used for calculating the fugitive emissions of the coal mining in China. Coal mining emission factors depend strongly on the type of coal extraction (underground mining emitting up to 10 times more than surface mining), the geological underground structure (very region-specific) and history (basin uplift), and the quality of the coal (brown coal emitting more than hard coal). The EDGARv4.2FT2012 seems to have overestimated by a factor of 2 the emission factor for the coal mining in China and allocated this to very few coal mine locations (hotspot emissions). A recent county-based inventory of Chinese methane emissions also confirms the overestimate of about >+38 % with total anthropogenic emissions estimated at 43 ± 6 Tg CH₄ yr^-1 (Peng et al., 2016). Also, assimilating also 13CH₄ data, Thompson et al. (2015)

showed that their prior (based on EDGARv4.2FT2010) overestimated the Chinese methane emissions by 30 %; however, they found no significant difference in the coal sector estimates between prior and posterior. EDGARv4.2 follows the IPCC guidelines 2006, which recommends region-specific data. However, the EDGARv4.2 inventory compilation used the European averaged emission factor for CH₄ from coal mine production in substitution for missing data, which seems to be twice too high in China. This highlights that significant errors on emission estimates may result from inappropriate use of some emission factor and that applying “Tier 1” for coal mine emissions is not accurate enough, as stated by the IPCC guidelines. The upcoming new version of EDGARv4.3.2 will revise this down and distribute the fugitive CH₄ from coal mining to more than 80 times more coal mining locations in China.

For the 2003–2012 decade, methane emissions from coal mining are estimated at 34 % of total fossil-fuel-related emissions of methane (41 Tg CH₄ yr^-1, range of 26–50), consistent with the AMAP report when considering the evolution since 2005. An additional very small source corresponds to fossil fuel fires (mostly underground coal fires, ∼ 0.1 Tg yr^-1, EDGARv4.2FT2010).

Natural gas is comprised primarily of methane, so any leaks during drilling of the wells, extraction, transportation, storage, gas distribution, and incomplete combustion of gas flares contribute to methane emissions (Lamb et al., 2015; Shorter et al., 1996). Fugitive permanent emissions (e.g. due to leaky valves and compressors) should be distinguished from intermittent emissions due to maintenance (e.g. purging and draining of pipes). During transportation, leakage can occur in gas transmission pipelines, due to corrosion, manufacturing, welding, etc. According to Lelieveld et al. (2005), the CH₄ leakage from gas pipelines should be relatively low; however, distribution networks in older cities have increased leakage, especially those with cast-iron and unprotected steel pipelines. Recent measurement campaigns in different cities in the USA and Europe also revealed that significant leaks occur in specific locations (e.g. storage facilities, city gates, well and pipeline pressurization/depressurization points) along the distribution networks to the end-users (Jackson et al., 2014a; McKain et al., 2015). However, methane emissions can vary a lot from one city to another depending in part on the age of city infrastructure (i.e. older cities on average have higher emissions). Ground movements (landslides, earthquakes, tectonic movements) can also release methane. Finally, additional methane emissions from the oil industry (e.g. refining) and production of charcoal are estimated to be a few Tg CH₄ yr^-1 only (EDGARv4.2, 2011). In many facilities, such as gas and oil fields, refineries and offshore platforms, venting of natural gas is now replaced by flaring with a partial conversion into CO₂; these two processes are usually considered together in inventories of oil and gas industries.

Methane emissions from oil and natural gas systems also vary greatly in different global inventories (46 to 98 Tg yr^-1 in 2005; Höglund-Isaksson et al., 2015). The inventories rely on the same sources and magnitudes regarding the activity data. Thus, the derived differences result from different methodologies and parameters used, including both emission and activity factors. Those factors are country or even site-specific, and the few field measurements available often combine oil and gas activities (Brandt et al., 2014) and remain largely unknown for most major oiland gas-producing countries. Depending on the country, the emission factors reported may vary by 2 orders of magnitude for oil production and by 1 order of magnitude for gas production (Table 5.5 of Höglund-Isaksson et al., 2015). The GAINS estimate of methane emissions from oil production is 4 times higher than EDGARv4.2FT2010 and USEPA. For natural gas, the uncertainty is also large (factor of 2), albeit smaller than for oil production. The difference in these estimates comes from the methodology used. Indeed, during oil extraction, the gas generated can be either recovered (re-injected or utilized as an energy source) or not recovered (flared or vented to the atmosphere). The recovery rates vary from one country to another (being much higher in the USA, Europe and Canada than elsewhere), and accounting for country-specific rates of generation and recovery of associated gas might lead to an amount of gas released into the atmosphere 4 times higher during oil production than when using default values (Höglund-Isaksson, 2012). This difference in methodology explains, in part, why GAINS estimates are higher than EDGARv4.2FT2010 and USEPA. Another challenge lies in determining the amount of flared or vented unrecovered gas, with venting emitting CH₄, whereas flaring converts all or most methane (often > 99 %) to CO₂. The balance of flaring and venting also depends on the type of oil: flaring is less common for heavy oil wells than conventional ones (Höglund-Isaksson et al., 2015). Satellite images can detect flaring (Elvidge et al., 2009, 2016) and may be used to verify the country estimates, but such satellites cannot currently be used to estimate the efficiency of CH₄ conversion to CO₂.

For the 2003–2012 decade, methane emissions from upstream and downstream natural oil and gas sectors are estimated to represent about 65 % of total fossil CH₄ emissions (79 Tg CH₄ yr^-1, range of 69–88, Table 2), with a lower uncertainty range than for coal emissions for most countries.

Production of natural gas from the exploitation of hitherto unproductive rock formations, especially shale, began in the 1980s in the US on an experimental or small-scale basis. Then, from early 2000s, exploitations started at large commercial scale. Two techniques developed and often applied together are horizontal drilling and hydraulic fracturing. The shale gas contribution to total natural gas production in the United States reached 40 % in 2012, growing rapidly from only small volumes produced before 2005 (EIA, 2015). Indeed, the practice of high-volume hydraulic fracturing (fracking) for oil and gas extraction is a growing sector of methane and other hydrocarbon production, especially in the US. Most recent studies (Miller et al., 2013; Moore et al., 2014; Olivier and Janssens-Maenhout, 2014; Jackson et al., 2014b; Howarth et al., 2011; Pétron et al., 2014; Karion et al., 2013) albeit not all (Allen et al., 2013; Cathles et al., 2012; Peischl et al., 2015) suggest that methane emissions are underestimated by inventories and agencies, including the USEPA. For instance, emissions in the Barnett Shale region of Texas from both bottom-up and top-down measurements showed that methane emissions from upstream oil and gas infrastructure were 90 % larger than estimates based on the USEPA’s inventory and corresponded to 1.5 % of natural gas production (Zavala-Araiza et al., 2015). This study also showed that a few high emitters, neglected in the inventories, dominated emissions. Moreover these high emitting points, located on the conventional part of the facility, could be avoided through better operating conditions and repair of malfunctions. It also suggests that emission factor of conventional and non-conventional gas facilities might not be as different as originally thought (Howarth et al., 2011). Field measurements suggest that emission factors for unconventional gas are higher than for conventional gas, though the uncertainty, largely site-dependent, is large, ranging from small leakage rate of 1–2 % (Peischl et al., 2015) to widely spread rates of 3–17 % (Caulton et al., 2014; Schneising et al., 2014). For current technology, the GAINS model has adopted an emission factor of 4.3 % for shale-gas mining, still awaiting a clear consensus across studies.

This category includes methane emissions related to livestock (enteric fermentation and manure), rice cultivation, landfills, and waste-water handling. Of all types of emission, livestock is by far the largest emitter of CH₄, followed by waste handling and rice cultivation. Field burning of agricultural residues was a minor source of CH₄ reported in emission inventories. The spatial distribution of methane emissions from agriculture and waste handling is presented in Fig. 3 based on the mean gridded maps provided by EDGARv4.2FT2010 and GAINS over the 2003– 2012 decade.

Global emissions for agriculture and waste are estimated at 195 Tg CH₄ yr^-1 (range 178–206, Table 2), representing 57 % of total anthropogenic emissions.

Domestic livestock such as cattle, buffalo, sheep, goats, and camels produce a large amount of methane by anaerobic microbial activity in their digestive systems (Johnson et al., 2002). A very stable temperature (39 ◦C), a stable pH (6.5– 6.8) in their rumen, and constant flow of plants (cattle graze many hours per day) induce a production of metabolic hydrogen, used by methanogenic Archaea together with CO₂ to produce methane. The methane and carbon dioxide are released from the rumen mainly through the mouth of multi-stomached ruminants (eructation, ∼ 87 % of emissions) or absorbed in the blood system. The methane produced in the intestines and partially transmitted through the rectum is only

∼ 13 %. There are about 1.4 billion cattle globally, 1 billion sheep, and nearly as many goats. The total number of animals is growing steadily (http://faostat3.fao.org), although the number is not linearly related to the CH₄ emissions they produce; emissions are strongly influenced by the total weight of the animals and their diet. Cattle, due to their large population, large size, and particular digestive characteristics, account for the majority of enteric fermentation CH₄ emissions from livestock, particularly, in the United States (USEPA, 2016). Methane emissions from enteric fermentation are also variable from one country to another as cattle experience water-limited conditions that highly vary spatially and temporally (especially in the tropics).

In addition, when livestock or poultry manure are stored or treated in systems that promote anaerobic conditions (e.g. as a liquid/slurry in lagoons, ponds, tanks, or pits), the decomposition of the volatile solids component in the manure tends to produce CH₄. When manure is handled as a solid (e.g. in stacks or dry lots) or deposited on pasture, range, or paddock lands, it tends to decompose aerobically and produce little or no CH₄. Ambient temperature, moisture, and manure storage or residency time affect the amount of CH₄ produced because they influence the growth of the bacteria responsible for CH₄ formation. For non-liquid-based manure systems, moist conditions (which are a function of rainfall and humidity) can promote CH₄ production. Manure composition, which varies with animal diet, growth rate, and type, including the animal’s digestive system, also affects the amount of CH₄ produced. In general, the greater the energy contents of the feed, the greater the potential for CH₄ emissions. However, some higher-energy feeds also are more digestible than lower quality forages, which can result in less overall waste excreted from the animal (USEPA, 2006).

In 2005, global methane emissions from enteric fermentation and manure are estimated in the range of 96– 114 Tg CH₄ yr^-1 in the GAINS model and USEPA inventory, respectively, and in the range of 98–105 Tg CH₄ yr^-1 suggested by Kirschke et al. (2013). They are consistent with the FAO-CH₄ estimate of 102 Tg CH₄ yr^-1 for 2005 (Tubiello et al., 2013).

Here, for the 2003–2012 decade, based on all the databases aforementioned, we infer a range of 97–111 Tg CH₄ yr^-1 for the combination of enteric fermentation and manure with a mean value of 106 Tg CH₄ yr^-1 (Table 2), about one-third of total global anthropogenic emissions.

Most of the world’s rice is grown on flooded fields (Baicich, 2013). Under these shallow-flooded conditions, aerobic decomposition of organic matter gradually depletes most of the oxygen in the soil, resulting in anaerobic conditions under which methanogenic Archaea decompose organic matter and produce methane. Most of this methane is oxidized in the underlying soil, while some is dissolved in the floodwater and leached away. The remaining methane is released to the atmosphere, primarily by diffusive transport through the rice plants, but also methane escapes from the soil via diffusion and bubbling through floodwaters (USEPA, 2016; Bridgham et al., 2013).

The water management systems used to cultivate rice are one of the most important factors influencing CH4 emissions and is one of the most promising approach to mitigate the CH4 emissions from rice cultivation (e.g. periodical drainage and aeration not only causes existing soil CH4 to oxidize but also inhibits further CH4 production in soils (Simpson et al., 1995; USEPA, 2016; Zhang et al., 2016). Upland rice fields are not flooded and, therefore, are not believed to produce much CH4. Other factors that influence CH4 emissions from flooded rice fields include fertilization practices (i.e. the use of urea and organic fertilizers), soil temperature, soil type (texture and aggregated size), rice variety and cultivation practices (e.g. tillage, seeding, and weeding practices) (USEPA, 2011, 2016; Kai et al., 2011; Yan et al., 2009; Conrad et al., 2000). For instance, methane emissions from rice paddies increase with organic amendments (Cai et al., 1997) but can be mitigated by applying other types of fertilizers (mineral, composts, biogas residues, wet seeding) (Wassmann et al., 2000). Some studies have suggested that decreases in microbial emissions, particularly due to changes in the practice of rice cultivation, could be responsible for a ∼ 15 Tg CH4 yr−1 decrease over the period from 1980s to 2000s (Kai et al., 2011).

The geographical distribution of the emissions is assessed by global (USEPA, 2006, 2012; EDGARv4.2FT2010, 2013) and regional (Peng et al., 2016; Chen et al., 2013; Chen and Prinn, 2006; Yan et al., 2009; Castelán-Ortega et al., 2014; Zhang et al., 2014) inventories or by land surface models (Spahni et al., 2011; Zhang and Chen, 2014; Ren et al., 2011; Tian et al., 2010, 2011; Li et al., 2005; Pathak et al., 2005). The emissions show a seasonal cycle, peaking in the summer months in the extratropics associated with the monsoon and land management. Similar to emissions from livestock, emissions from rice paddies are influenced not only by extent of rice field area (equivalent to the number of livestock) but also by changes in the productivity of plants as these alter the CH₄ emission factor used in inventories.

The largest emissions are found in Asia (Hayashida et al., 2013), with China (5–11 Tg CH₄ yr^-1; Chen et al., 2013; Zhang et al., 2016) and India (∼ 3–5 Tg CH₄ yr^-1; Bhatia et al., 2013) accounting for 30 to 50 % of global emissions (Fig. 3). The decrease of CH₄ emissions from rice cultivation over the past decades is confirmed in most inventories, because of the decrease in rice cultivation area, the change in agricultural practices, and a northward shift of rice cultivation since 1970s (e.g. Chen et al., 2013). Furthermore, recent studies revealed that, together, high carbon dioxide concentrations and warmer temperatures predicted for the end of the twenty-first century will about double the amount of methane emitted per kilogramme of rice produced (van Groenigen et al., 2013).

Based on global inventories only, global methane emissions from rice paddies are estimated in the range 24–36 Tg CH₄ yr−1 for the 2003–2012 decade, with a mean value of 30 Tg CH₄ yr^-1 (Table 2), about 9 % of total global anthropogenic emissions. The lower estimate (24 Tg CH₄ yr^-1) is provided by FAO-CH₄ inventory (Tubiello et al., 2013), which is based on a mix of FAO statistics for crop production and IPCC guidelines.

This category includes all the combustion processes: biomass (forests, savannahs, grasslands, peats, agricultural residues) and biofuels in the residential sector (stoves, boilers, fireplaces). Biomass and biofuel burning emits methane under incomplete combustion conditions, when oxygen availability is insufficient such as charcoal manufacture and smouldering fires. The amount of methane that is emitted during the burning of biomass depends primarily on the amount of biomass, the burning conditions, and the material being burned. At the global scale, biomass and biofuel burning lead to methane emissions of 27–35 Tg CH₄ yr^-1 with an average of 30 Tg CH₄ yr^-1 (2003–2012 decade, Table 2), of which 30–50 % is biofuel burning (Kirschke et al., 2013).

In this study, we use the large-scale biomass burning (forest, savannah, grassland and peat fires) from specific biomass burning inventories and the biofuel burning contribution for the inventories (USEPA, GAINS and EDGAR).

The spatial distribution of methane emissions from biomass burning over the 2003–2012 decade is presented in Fig. 3 and is based on the mean gridded maps provided by EDGARv4.2FT2010 and GAINS for the biofuel burning, and based on the mean gridded maps provided by the biomass burning inventories presented thereafter.

Biomass that is used to produce energy for domestic, industrial, commercial, or transportation purposes is hereafter called biofuel burning. A largely dominant fraction of methane emissions from biofuels comes from domestic cooking or heating in stoves, boilers and fireplaces, mostly in open cooking fires where wood, charcoal, agricultural residues or animal dung are burnt. Although more than 2 billion people, mostly in developing and emerging countries, use solid biofuels to cook and heat their homes on a daily basis (André et al., 2014), methane emissions from biofuel combustion have not yet received the attention it should have to estimate its magnitude. Other much smaller contributors include agricultural burning (∼ 1–2 Tg yr^-1) and road transportation (< 1 Tg yr^-1). Biofuel burning estimates are gathered from USEPA, GAINS and EDGAR inventories.

In this study, biofuel burning is estimated to contribute 12 Tg CH₄ yr^-1 [10–14] to the global methane budget, about 3 % of total global anthropogenic emissions.

Natural methane sources include wetland emissions as well as emissions from other land water systems (lakes, ponds, rivers, estuaries), land geological sources (seeps, microseepage, mud volcanoes, geothermal zones, and volcanoes, marine seepages), wild animals, wildfires, termites, terrestrial permafrost and oceanic sources (geological and biogenic). Many sources have been recognized but their magnitude and variability remain uncertain (USEPA, 2010a; Kirschke et al., 2013).

Wetlands are generally defined as ecosystems in which water saturation or inundation (permanent or not) dominates the soil development and determines the ecosystem composition (USEPA, 2010a). Such a broad definition needs to be refined when it comes to methane emissions. In this work, we define wetlands as ecosystems with inundated or saturated soils where anaerobic conditions lead to methane production (USEPA, 2010a; Matthews and Fung, 1987). This includes peatlands (bogs and fens), mineral wetlands (swamps and marshes), and seasonal or permanent floodplains. It excludes exposed water surfaces without emergent macrophytes, such as lakes, rivers, estuaries, ponds, and dams (addressed in the next section), as well as rice agriculture (see Sect. 3.1.4., rice cultivation paragraph). Even with this definition, one can consider that part of the wetlands could be considered as anthropogenic systems, being affected by human-driven land-use changes (Woodward et al., 2012). In the following we keep the generic denomination wetlands for natural and human-influenced wetlands.

A key feature of wetland systems producing methane is anaerobic soils, where high water table or flooded conditions limit oxygen availability and create conditions for methanogenesis. In anoxic conditions, organic matter can be degraded by methanogens that produce CH₄. The three most important factors influencing methane production in wetlands are the level of anoxia (linked to water table), temperature and substrate availability (Wania et al., 2010; Valentine et al., 1994; Whalen, 2005). Once produced, methane can reach the atmosphere through a combination of three processes: molecular diffusion, plant-mediated transport and ebullition. On its way to the atmosphere, methane can be partly or completely oxidized by a group of bacteria, called methanotrophs, which use methane as their only source of energy and carbon (USEPA, 2010a). Concurrently, methane from the atmosphere can diffuse into the soil column and be oxidized (see Sect. 3.3.4).

Land surface models estimate CH₄ emissions through a series of processes, including CH₄ production, CH₄ oxidation and transport and are further regulated by the changing environmental factors (Tian et al., 2010; Xu et al., 2010; Melton et al., 2013). In these models, methane emissions from wetlands to the atmosphere are computed as the product of an emission density (which can be negative; mass per unit area and unit time) multiplied by a wetland extent (see the model intercomparison studies by Melton et al., 2013, and Bohn et al., 2015). The CH₄ emission density is represented in land surface models with varying levels of complexity. Many models link CH₄ emission to net primary production (NPP) though production of exudates or litter and soil carbon to yield heterotrophic respiration estimates. A proportion of the heterotrophic respiration estimate is then taken to be CH₄ production (Melton et al., 2013). The oxidation of produced (and becoming atmospheric) methane in the soil column is then either represented explicitly (e.g. Riley et al., 2011; Grant and Roulet, 2002), or just fixed proportionally to the production (Wania et al., 2013).

In land surface models, wetland extent is either prescribed (from inventories or remote-sensing data) or computed using hydrological models accounting for the fraction of grid cell with flat topography prone to high water table (e.g. Stocker et al., 2014; Kleinen et al., 2012), or from data assimilation against remote-sensed observations (Riley et al., 2011). Mixed approaches can also be implemented with tropical extent prescribed from remote sensing and northern peatland extent explicitly computed (Melton et al., 2013). Wetland extent appears to be a large contributor to uncertainties in methane emissions from wetlands (Bohn et al., 2015). For instance, the maximum wetland extent on a yearly basis appeared to be very different among land surface models in Melton et al. (2013), ranging from 7 to 27 Mkm2.Passive and active remote-sensing data in the microwave domain have been used to retrieve inundated areas, as with the Global Inundation Extent from Multi-Satellites product (GIEMS, Prigent et al., 2007; Papa et al., 2010). These remote-sensed data do not exactly correspond to wetlands, as all flooded areas are not wetlands (in methane emission sense) and some wetlands (e.g. northern bogs) are not always flooded. Inundated areas also include inland water bodies (lakes, ponds, estuaries) and rice paddies, which have to be filtered out to compute wetland emissions. Overall, current remote sensing of wetlands tends to underestimate wetland extent partly because of signal deterioration over dense vegetation and partly because microwave signals only detect water above or at the soil surface and therefore do not detect emitting peatlands that are not inundated (Prigent et al., 2007). For example, the Global Lakes and Wetlands Dataset (GLWD) (Lehner and Döll, 2004) estimates between 8.2 and 10.1 Mkm2 of wetlands globally, while remote-sensing inundation area is smaller, i.e. ∼ 6 Mkm2(Prigent et al., 2007). Some ancillary data used in the GIEMS processing are not available after 2007 and have prevented so far the extension of the dataset after 2007.

Integrated at the global scale, wetlands are the largest and most uncertain source of methane to the atmosphere (Kirschke et al., 2013). An ensemble of land surface models estimated the range of methane emissions of natural wetlands

at 141–264 Tg CH₄ yr^-1 for the 1993–2004 period, with a mean and 1σ value of 190 ± 39 Tg CH₄ yr^-1 (Melton et al., 2013). Kirschke et al. (2013) assessed a consistently large emission range of 142–287 Tg CH₄ yr^-1, using the Melton et al. (2013) land surface models and atmospheric inversions. These emissions represent about 30 % of the total methane source. The large range in the estimates of wetland emissions results from difficulties in defining wetland CH₄producing areas as well as in parameterizing terrestrial anaerobic sources and oxidative sinks (Melton et al., 2013; Wania et al., 2013).

In this work, following Melton et al. (2013), 11 land surface models (Table 1) computing net CH₄ emissions have been run under a common protocol with a 30-year spin-up (1901–1930) followed by a simulation until the end of 2012 forced by CRU-NCEP v4.0 reconstructed climate fields. Atmospheric CO2 influencing NPP was also prescribed in the models, allowing the models to separately estimate carbon availability for methanogenesis. In all models, the same wetland extent (SWAMPS-GLWD) has been prescribed. The SWAMPS-GLWD is a monthly global wetland area dataset, which has been developed to overcome the aforementioned issues and combines remote-sensing data from Schroeder et al. (2015) and GLWD inventory in order to develop a monthly global wetland area dataset (Poulter et al., 2016). Briefly, GLWD was used to set the annual mean wetland area, to which a seasonal cycle of fractional surface water was added using data from the Surface WAter Microwave Product Series Version 2.0 (SWAMPS) (Schroeder et al., 2015). The combined GLWD-SWAMPS product leads to a maximum annual wetland area of 10.5 Mkm2 (8.7 Mkm2on average, about 5.5 % of than global land surface). The largest wetland areas in the SWAMPS-GLWD are in Amazonia, the Congo Basin, and the western Siberian lowlands, which in previous studies have appeared to be strongly underestimated by several inventories (Bohn et al., 2015). However, wetlands above 70◦ N appear under-represented in GLWD as compared to Sheng et al. (2004) and Peregon et al. (2008). Indeed, approximately half of the global natural wetland area lies in the boreal zone between 50 and 70◦ N, while 35 % can be found in the tropics, between 20◦ N and 30◦ S (Matthews and Fung, 1987; Aselmann and Crutzen, 1989). Despite the lower area extent, the higher per-unit area methane emissions of tropical wetlands results in a larger wetland source from the tropics than from the boreal zone (Melton et al., 2013).

The average emission map from wetlands for 2003–2012 built from the 11 models is plotted in Fig. 3. The zones with the largest emissions reflect the GLWD database: the Amazon basin, equatorial Africa and Asia, Canada, western Siberia, eastern India, and Bangladesh. Regions where methane emissions are robustly inferred (i.e. regions where mean flux is larger than the standard deviation of the models) represent 80 % of the total methane flux due to natural wetlands. Main primary emission zones are consistent between models, which is clearly favoured by the common wetland extend prescribed. But still, the different sensitivity of the models to temperature can generate substantial different patterns, such as in India. Some secondary (in magnitude) emission zones are also consistently inferred between models: Scandinavia, continental Europe, eastern Siberia, central USA, and tropical Africa. Using improved regional methane emission datasets (such as studies over North America, Africa, China, and Amazon) can enhance the accuracy of the global budget assessment (Tian et al., 2011; Xu and Tian, 2012; Ringeval et al., 2014; Valentini et al., 2014).

The resulting global flux range for natural wetland emissions is 153–227 Tg CH₄ yr^-1 for the 2003–2012 decade, with an average of 185 Tg CH₄ yr^-1 with a 1σ standard deviation of 21 Tg CH₄ yr^-1 (Table 2).

This category includes methane emissions from freshwater systems (lakes, ponds, rivers) and from brackish waters of estuaries. Methane emissions from fresh waters and estuaries occur through a number of pathways including (1) continuous or episodic diffusive flux across water surfaces, (2) ebullition flux from sediments, (3) flux mediated through the aerenchyma of emergent aquatic macrophytes (plant transport) in littoral environments, and also for reservoirs, (4) degassing of CH₄ in the turbines, and (5) elevated diffusive emissions in rivers downstream of the turbines especially if water through the turbines is supplied from anoxic CH₄-rich water layers in the reservoir (Bastviken et al., 2004; Guérin et al., 2006, 2016). It is very rare that complete emission budgets include all these types of fluxes. For methodological reasons many past and present flux measurements only account for the diffusive flux based on short-term flux chamber measurements where non-linear fluxes were often discarded. At the same time, diffusive flux is now recognized as a relatively small flux component in many lakes, compared to ebullition and plant fluxes (in lakes with substantial emergent macrophyte communities). The two latter fluxes are very challenging to measure, both typically being associated with shallow near-shore waters and having high spatiotemporal variability. Ebullition can also occur more frequently in areas with high sediment organic matter load and is by nature episodic with very high fluxes occurring over time frames of seconds followed by long periods without ebullition.

Freshwater contributions from lakes were first estimated to emit 1–20 Tg CH₄ yr^-1 based on measurements in two systems (Great Fresh Creek, Maryland, and Lake Erie; Ehhalt, 1974). A subsequent global emission estimate was 11– 55 Tg CH₄ yr^-1 based on measurements from three arctic lakes and a few temperate and tropical systems (Smith and Lewis, 1992), and 8–48 Tg CH₄ yr^-1 using extended data from all of the lake rich biomes (73 lakes; Bastviken et al., 2004). Combining results from Bastviken et al. (2004) and Bastviken et al. (2011), Kirschke et al. (2013) reported a range of 8–73 Tg CH₄ yr^-1. Gradually, methane emissions from reservoirs and rivers have also been included in the most recent global estimate from fresh waters of 103 Tg CH₄ yr^-1, including emissions from non-saline lakes, reservoirs, ponds and rivers (data from 473 systems; Bastviken et al., 2011). Improved stream and river emission estimates of 27 Tg CH₄ yr^-1 were recently suggested (Stanley et al., 2016). Importantly, the previous estimates of inland water fluxes are not independent. Instead they represent updates from increasing data quantity and quality. It should also be noted that issues regarding spatiotemporal variability are not considered in consistent ways at present (Wik et al., 2016a; Natchimuthu et al., 2015).

Present data do not allow for separating inland water fluxes over the different time periods investigated in this paper. The global estimates provided are therefore assumed to be constant for this study. Here we combine the latest estimates of global freshwater CH₄ emissions (Bastviken et al., 2011) with a more recent regional estimate for latitudes above 50◦ N at present (Wik et al., 2016b) and new extrapolations for tropical river emissions (Borges et al., 2015; Sawakuchi et al., 2014) and streams (Stanley et al., 2016). High-latitude lakes include both post-glacial lakes and thermokarst lakes (water bodies formed by thermokarst), the latter having larger emissions per square metre but smaller regional emissions than the former because of smaller areal extent (Wik et al., 2016b). Water body depth, sediment type, and ecoclimatic region are the key factors explaining variation in methane fluxes from lakes (Wik et al., 2016b).

Altogether, these studies consider data from more than 900 systems, of which ∼ 750 are located north of 50◦ N. In this context we only consider fluxes from open waters assuming that plant-mediated fluxes are included in the wetland emission term. The average total estimated open water emission including the recent estimates from smaller streams is 122 Tg CH₄ yr^-1. The uncertainty is high with a coefficient of variation ranging from 50 to > 100 % for various flux components and biomes (Bastviken et al., 2011) resulting in a minimum uncertainty range of 60–180 Tg CH₄ yr^-1. The present data indicate that lakes or natural ponds, reservoirs, and streams/rivers account for 62, 16 and 22 % of the average fluxes, respectively (given the large uncertainty the percentages should be seen as approximate relative magnitudes only).

Potentially, the emissions from reservoirs should be allocated to anthropogenic emissions (not done here). Regarding lakes and reservoirs, tropical (< 30◦ latitude) and temperate (30–50◦ latitude) emissions represent 49 and 33 % of the flux, respectively, with 18 % left for regions above 50◦ latitude. For comparison, approximately 40 % of the inland water surface area is found above 50◦ latitude in the Northern Hemisphere and 34 % of the area is situated between 20◦ S and 20◦ N (Verpoorter et al., 2014). Ebullition typically accounted for 50 to more than 90 % of the flux from the water bodies, while contributions from ebullition appear lower from rivers, although this is currently debated (e.g. Crawford et al., 2014). Several aspects will need consideration to reduce the remaining uncertainty in the freshwater fluxes, including the generation of flux measurement that is more representative in time and space and an update of global lake area databases (e.g. GLOWAB, Verpoorter et al., 2014).

Significant amounts of methane, produced within the Earth’s crust, naturally migrate to the atmosphere through tectonic faults and fractured rocks. Major emissions are related to hydrocarbon production in sedimentary basins (microbial and thermogenic methane), through continuous exhalation and eruptions from onshore and shallow marine gas/oil seeps and through diffuse soil microseepage (after Etiope, 2015). Specifically, six source categories have been considered. Five are onshore sources: mud volcanoes (sedimentary volcanism), gas and oil seeps (independent of mud volcanism), microseepage (diffuse exhalation from soil in petroleum basins), geothermal (non-volcanic) manifestations and volcanoes. One source is offshore: submarine seepage (several types of gas manifestation at the seabed). Figure 4a shows the areas and locations potentially emitting geological methane, showing diffuse potential microseepage regions, macroseepage locations (oil–gas seeps, mud volcanoes) and geothermal/volcanic areas (built from Etiope, 2015), which represent more than 1000 emitting spots.

Studies since 2000 have shown that the natural release to the Earth’s surface of methane of geological origin is an important global greenhouse gas source (Etiope and Klusman, 2002; Kvenvolden and Rogers, 2005; Etiope et al., 2008; USEPA, 2010a; Etiope, 2012, 2015). Indeed, the geological source is in the top-three natural methane sources after wetlands (and with freshwater systems) and about 10 % of total methane emissions, of the same magnitude or exceeding other sources or sinks, such as biomass burning, termites and soil uptake, considered in recent IPCC assessment reports (Ciais et al., 2013).

In this study, the following provided estimates were derived by bottom-up approaches based on (a) the acquisition of thousands of land-based flux measurements for various seepage types in many countries, and (b) the application of the same procedures typically used for natural and anthropogenic gas sources, following upscaling methods based on the concepts of “point sources”, “area sources”, “activity” and “emission factors”, as recommended by the air pollutant emission guidebook of the European Environment Agency (EMEP/EEA, 2009). Our estimate is consistent with a top-down global verification, based on observations of radiocarbon-free (fossil) methane in the atmosphere (Etiope et al., 2008; Lassey et al., 2007b), with a range of 33–75 Tg CH₄ yr^-1.

As a result, in this study, the global geological methane emission is estimated in the range of 35–76 Tg CH₄ yr^-1 (mean of 52 Tg CH₄ yr^-1), with 40 Tg CH₄ yr−1 [30–56] for onshore emissions (10–20 Tg CH₄ yr^-1 for mud volcanoes, 3–4 Tg yr^-1 for gas–oil seeps, 10–25 Tg yr^-1 for microseepage, 2–7 Tg CH₄ yr^-1 for geothermal/volcanic manifestations) and 12 Tg CH₄ yr^-1 [5–20] for offshore emissions through marine seepage (Rhee et al., 2009; Berchet et al., 2016; Etiope, 2012; see Sect. 3.2.6 for offshore contribution explanations).

Termites are important decomposer organisms, which play a very relevant role in the cycling of nutrients in tropical and subtropical ecosystems (Sanderson, 1996). The degradation of organic matter in their gut, by symbiotic anaerobic microorganisms, leads to the production of CH₄ and CO2 (Sanderson, 1996). The upscaling approaches which have been used to quantify the contribution of termites to global CH₄ emissions (Sanderson, 1996; Sugimoto et al., 1998; Bignell et al., 1997) are affected by large uncertainties, mainly related to the effect of soil and mound environments on net CH₄ emissions; the quantification of termite biomass for each ecosystem type; and the impact of land-use change on termite biomass. For all these factors, uncertainty mainly comes from the relatively small number of studies compared to other CH₄ sources. In Kirschke et al. (2013) (see their Supplement), a reanalysis of CH₄ emissions from termites at the global scale was proposed and CH₄ emissions per unit of surface were estimated as the product of termite biomass, termite CH₄ emissions per unit of termite mass and a scalar factor expressing the effect of land-use/land-cover change. The latter two terms were estimated from published literature reanalysis (Kirschke et al., 2013, Supplement). A climate zoning (following the Köppen–Geiger classification) was applied to updated climate datasets by Santini and Di Paola (2015) and was adopted to take into account different combinations of termite biomass per unit area and CH₄ emission factor per unit of termite biomass. In the case of tropical climate, first termites’ biomass was estimated by a simple regression model representing its dependence on gross primary productivity (Kirschke et al., 2013, Supplement), whereas termites’ biomass for forest and grassland ecosystems of the warm temperate climate and for shrublands of the Mediterranean subclimate were estimated from data reported by Sanderson (1996). CH₄ emission factor per unit of termite biomass was derived from published literature and was estimated equal to 2.8 mg CH₄ g^-1 termite h^-1 for tropical ecosystems and Mediterranean shrublands (Kirschke et al., 2013) and 1.7 mg CH₄ g^-1 termite h^-1 for temperate forests and grasslands (Fraser et al., 1986). Emissions were scaled up in GIS environment and annual CH₄ fluxes computed for the three periods 1982–1989, 1990–1999 and 2000–2007 representative of the 1980s, 1990s and 2000s, respectively. CH₄ emissions showed only little interannual and interdecadal variability (0.1 Tg CH₄ yr^-1) and strong regional variability with tropical South America and Africa being the main sources (36 and 30 % of the global total emissions, respectively) due to the extent of their natural forest and savannah ecosystems (Fig. 4b). For the 2000s, a global total of 8.7 ± 3.1 Tg CH₄ yr^-1 (range 3–15 Tg CH₄ yr^-1) was obtained. This value is close to the average estimate derived from previous upscaling studies which report values spanning from 2 to 22 Tg CH₄ yr^-1 (Ciais et al., 2013). In this study, we adopt a value of 9 Tg CH₄ yr−1 (range 3–15 Tg CH₄ yr^-1, Table 2).

As for domestic ruminants, wild ruminants eruct or exhale methane through the microbial fermentation process occurring in their rumen (USEPA, 2010a). Global emissions of CH₄ from wild animals range from 2–6 Tg CH₄ yr^-1 (Leng, 1993) to 15 Tg CH₄ yr^-1 (Houweling et al., 2000). The global distribution of CH₄ emissions from wild ruminants is generally estimated as a function of the percentage and type of vegetation consumed by the animals (Bouwman et al., 1997). However, as suspected, numerous and various wild animals live partly hidden in the forests, savannahs, etc., challenging the assessment of these emissions.

The range adopted in this study is 2–15 Tg CH₄ yr^-1 with a mean value of 10 Tg CH₄ yr^-1 (Table 2).

Possible sources of oceanic CH₄ include the following: (1) leaks from geological marine seepage (see also Sect. 3.2.3); (2) production from sediments or thawing subsea permafrost; (3) emission from the destabilization of marine hydrates and (4) in situ production in the water column, especially in the coastal ocean because of submarine groundwater discharge (USEPA, 2010a). Once at seabed, methane can be transported through the water column by diffusion in a dissolved form (especially in the upwelling zones) or by ebullition (gas bubbles, e.g. from geological marine seeps), for instance, in shallow waters of continental shelves. Among these different origins of oceanic methane, hydrates have attracted a lot of attention. Methane hydrates (or sometimes called clathrates) are ice-like crystals formed under specific temperature and temperature conditions (Milkov, 2005). The stability zone for methane hydrates (high pressure, ambient temperatures) can be found in the shallow lithosphere (i.e. < 2000 m depth), either in the continental sedimentary rocks of polar regions or in the oceanic sediments at water depths greater than 300 m (continental shelves, sediment– water interface) (Kvenvolden and Rogers, 2005; Milkov, 2005). Methane hydrates can be either of biogenic origin (formed in situ at depth in the sediment by microbial activity) or of thermogenic origin (non-biogenic gas migrated from deeper sediments and trapped due to pressure/temperature conditions or due to some capping geological structure such as marine permafrost). The total stock of marine methane hydrates is large but uncertain, with global estimates ranging from hundreds to thousands of Pg CH₄ (Klauda and Sandler, 2005; Wallmann et al., 2012).

If the production of methane at seabed can be of importance, for instance, marine seepages emit up to 65 Tg CH₄ yr^-1 globally at seabed level (USEPA, 2010a); more uncertain is the flux of oceanic methane reaching the atmosphere. For example, bubble plumes of CH₄ from the seabed have been observed in the water column but not detected in the Arctic atmosphere (Westbrook et al., 2009; Fisher et al., 2011). A large part of the seabed CH₄ production and emission is oxidized in the water column and does not reach the atmosphere (James et al., 2016). There are several barriers preventing methane from being expelled to the atmosphere. From the bottom to the top, gas hydrates and permafrost serve as a barrier to fluid and gas migration towards the seafloor (James et al., 2016). First, on centennial to millennium timescales, trapped gases may be released when permafrost is perturbed and cracks or through Pingolike features. At present, microbial processes are the most important control on methane emissions from marine environments. Aerobic oxidation in the water column is a very efficient sink, which allows very little methane even from established and vigorous gas seep areas or even gas well blowouts such as the Deepwater Horizon from reaching the atmosphere. Anaerobic methane oxidation, first described by Reeburgh and Heggie (1977), coupled to sulfate reduction controls methane losses from sediments to the overlying water (Reeburgh, 2007). Methane only escapes marine sediments in significant amounts from rapidly accumulating sedimentary environments or via advective processes such as ebullition or groundwater flow in shallow shelf regions. Anaerobic methane oxidation was recently demonstrated to be able to keep up with the thaw front of thawing permafrost in a region that had been inundated within the past 1000 years (Overduin et al., 2015). Second, the oceanic pycnocline is a physical barrier limiting the transport of methane (and other species) towards the surface. Third, another important mechanism stopping methane from reaching the ocean surface is the dissolution of bubbles into the ocean water. Although bubbling is the most efficient way to transfer methane from the seabed to the atmosphere, the fraction of bubbles actually reaching the atmosphere is very uncertain and critically depends on emission depths (< 100–200 m, McGinnis et al., 2015) and on the size of the bubbles (> 5–8 mm; James et al., 2016). Finally, surface oceans are aerobic and contribute to the oxidation of dissolved methane (USEPA, 2010a). However, surface waters can be more supersaturated than the underlying deeper waters, leading to a methane paradox (Sasakawa et al., 2008). Possible explanations involve upwelling in areas with surface mixed layers covered by sea ice (Damm et al., 2015) or methane produced within the anoxic centre of sinking particles (Sasakawa et al., 2008), but more work is needed to correct such an apparent paradox.

Permafrost is defined as frozen soil, sediment, or rock having temperatures at or below 0 ◦C for at least two consecutive years (ACIA, 2005; Arctic Research Commission, 2003). The total extent of permafrost zones of the Northern Hemisphere is about 15 % of the land surface, with values around 15 million square kilometres (Slater and Lawrence, 2013; Levavasseur et al., 2011; Zhang et al., 1999). Where soil temperatures have passed the 0 ◦C mark, thawing of the permafrost at its margins occurs, accompanied by a deepening of the active layer (Anisimov and Reneva, 2006) and possible formation of thermokarst lakes (Christensen et al., 2015). A total of 1035 ± 150 Pg of carbon can be found in the upper 3 m or permafrost regions, or ∼ 1300 Pg of carbon (1100 to 1500) Pg C for all permafrost (Hugelius et al., 2014; Tarnocai et al., 2009). The thawing permafrost can generate direct and indirect methane emissions. Direct methane emissions rely on the release of the methane contained in the thawing permafrost.

A series of recent studies define three distinct pathways for the production and emission of methane by living vegetation. First, plants produce methane through an abiotic photochemical process induced by stress (Keppler et al., 2006). This pathway was criticized (e.g. Dueck et al., 2007; Nisbet et al., 2009), and although numerous studies have since confirmed aerobic emissions from plants and better resolved its physical drivers (Fraser et al., 2015), global estimates still vary by 2 orders of magnitude (Liu et al., 2015) meaning any potential implication for the global methane budget remains highly uncertain. Second, plants act as “straws”, drawing methane produced by microbes in anoxic soils (Rice et al., 2010; Cicerone and Shetter, 1981). Third, the stems of living trees commonly provide an environment suitable for microbial methanogenesis (Covey et al., 2012). Static chambers demonstrate locally significant through-bark flux from both soil-based (Pangala et al., 2013, 2015), and tree-stem-based methanogens (Wang et al., 2016). These studies indicate trees are a significant factor regulating ecosystem flux; however, estimates of biogenic plant-mediated methane emissions at broad scales are complicated by overlap with methane consumption in upland soil and production in wetlands. Integrating plant-mediated emissions in the global methane budget will require untangling these processes to better define the mechanisms, spatio-temporal patterns, and magnitude of these pathways.

Methane is the most abundant reactive trace gas in the troposphere and its reactivity is important to both tropospheric and stratospheric chemistry. The main atmospheric sink of methane is its oxidation by the hydroxyl radical (OH), mostly in the troposphere, which contributes about 90 % of the total methane sink (Ehhalt, 1974). Other losses are by photochemistry in the stratosphere (reactions with chlorine atoms, Cl, and atomic oxygen, O(¹D)), by oxidation in soils (Curry, 2007; Dutaur and Verchot, 2007), and by photochemistry in the marine boundary layer (reaction with Cl; Allan et al., 2007; Thornton et al., 2010). Uncertainties in the total sink of methane as estimated by atmospheric chemistry models are of the order of 20–40 % (Kirschke et al., 2013). It is much less (10–20 %) when using atmospheric proxy methods (e.g. methyl chloroform, see below) as in atmospheric inversions (Kirschke et al., 2013). Methane is a significant source of water vapour in the middle to upper stratosphere and influences stratospheric ozone concentrations by converting reactive chlorine to less reactive hydrochloric acid (HCl). In the present release of the global methane budget, we essentially rely on the former analysis of Kirschke et al. (2013) and IPCC AR5. Following the ACCMIP model intercomparison (Lamarque et al., 2013), the ongoing Climate Chemistry Model Initiative (CCMI) and the upcoming Aerosols Chemistry Modeling Intercomparison Project (AerChemMIP) should allow obtaining updated estimates on methane chemical sinks and lifetimes.

OH radicals are produced following the photolysis of ozone (O3)in the presence of water vapour. OH is destroyed by reactions with CO, CH₄, and non-methane volatile organic compounds, but since OH exists in photochemical equilibrium with HO₂, the net effect of CH₄ oxidation on the HOx budget also depends on the level of NOx(Lelieveld et al., 2002) and other competitive oxidants. Considering its very short lifetime (a few seconds, Lelieveld et al., 2004), it is not possible to estimate global OH concentrations directly from observations. Observations are generally carried out within the boundary layer, while the global OH distribution and variability are more influenced by the free troposphere (Lelieveld et al., 2016). A series of experiments were conducted by several chemistryclimate models and chemistry transport models participating in the Atmospheric Chemistry and Climate Model Intercomparison Project (ACCMIP) to study the long-term changes in atmospheric composition between 1850 and 2100 (Lamarque et al., 2013). For the year 2000, the multimodel mean (14 models) global mass-weighted OH tropospheric concentration is 11.7 ± 1.0 × 105 molec cm−3(range 10.3–13.4 × 105 molec cm−3, Voulgarakis et al., 2013), consistent with the estimates of Prather et al. (2012) at 1.2 ± 1.3 × 105 molec cm−3. However, it is worth noting that, in the ACCMIP estimations, the differences in global

OH are larger between models than between pre-industrial, present and future emission scenario simulations. Indeed Lelieveld et al. (2016) suggest that tropospheric OH is buffered against potential perturbations from emissions, mostly due to chemistry and transport connections in the free troposphere, through transport of oxidants such as ozone. Besides the uncertainty on global OH concentrations, the OH distribution is highly discussed. Models are often high biased in the Northern Hemisphere leading to a NH / SH OH ratio greater than 1 (Naik et al., 2013). A methane inversion using a NH / SH OH ratio higher than 1 infers higher methane emissions in the Northern Hemisphere and lower in the tropics and in the Southern Hemisphere (Patra et al., 2014). However, there is recent evidence for parity in interhemispheric OH concentrations (Patra et al., 2014), which needs to be confirmed by other observational and modelderived estimates.

OH concentrations and their changes can be sensitive to climate variability (e.g. Pinatubo eruption, Dlugokencky et al., 1996), to biomass burning (Voulgarakis et al., 2015) and to anthropogenic activities. For instance, the recent increase of the oxidizing capacity of the troposphere in South and East Asia, associated with increasing NOxemissions and decreasing CO emissions (Mijling et al., 2013; Yin et al., 2015), possibly enhances CH₄ consumption and therefore limits the atmospheric impact of increasing emissions (Dalsøren et al., 2009). Despite such large regional changes, the global mean OH concentration was suggested to have changed only slightly over the past 150 years (Naik et al., 2013). This is due to the concurrent increases of positive influences on OH (water vapour, tropospheric ozone, nitrogen oxides (NOx ) emissions, and UV radiation due to decreasing stratospheric ozone) and of OH sinks (methane burden, carbon monoxide and non-methane volatile organic compound emissions and burden). However the sign and integrated magnitude (from 1850 to 2000) of OH changes is uncertain, varying from −13 to >+15 % among the ACCMIP models (mean of −1 %, Naik et al., 2013). Dentener et al. (2003) found a positive trend in global OH concentrations of 0.24 ± 0.06 % yr^-1 between 1979 and 1993, mostly explained by changes in the tropical tropospheric water vapour content. Accurate methyl chloroform atmospheric observations together with estimates of its emissions (Montzka and Fraser, 2003) allow an estimate of OH concentrations and changes in the troposphere from the 1980s. Montzka et al. (2011) inferred small interannual OH variability and trends (typical OH changes from year to year of less than 3 %) and attributed previously estimated large year-to-year OH variations before 1998 (e.g. Bousquet et al., 2005; Prinn et al., 2001) to overly large sensitivity of OH concentrations inferred from methyl chloroform measurements to uncertainties in the latter’s emissions. However, Prinn et al. (2005) also showed lower post-1998 OH variability that they attributed to the lack of strong post-1998 El Niño events. For the ACCMIP models providing continuous simulations over the past decades, OH interannual variability ranged from 0.4 to 0.9 %, consistent but lower than the value deduced from methyl chloroform measurements. However these runs take into account meteorology variability but not emission interannual variability (e.g. from biomass burning) and thus are expected to simulate lower OH interannual variability than in reality. As methyl chloroform has reached very low concentrations in the atmosphere, in compliance with the application of the Montreal Protocol and its amendments, a replacement compound is needed to estimate global OH concentrations. Several hydrochlorofluorocarbons and hydrofluorocarbons have been tested (Miller et al., 1998; Montzka et al., 2011; Huang and Prinn, 2002) to infer OH but do not yet provide equivalent results to methyl chloroform.

We report here a climatological range of 454– 617 Tg CH₄ yr^-1 as in Kirschke et al. (2013) for the total tropospheric loss of methane by OH oxidation in the 2000s.

Approximately 60 Tg CH₄ yr−1 enters the stratosphere by cross-tropopause mixing and the Hadley circulation (Reeburgh, 2007). Stratospheric CH₄ distribution is highly correlated to the changes in the Brewer–Dobson circulation (Holton, 1986) and may impact Arctic air through subsidence of isotopically heavy air depending on the polar vortex location (Röckmann et al., 2011). In the stratosphere, currently approximately 51 [16–84] Tg CH₄ yr^-1 (i.e. about 10 [3–16] % of the total chemical loss in the atmosphere) is lost through reactions with excited atomic oxygen O(¹D), atomic chlorine (Cl), atomic fluorine (F) and OH (Voulgarakis et al., 2013; Williams et al., 2012). The fraction of the stratospheric loss due to the different oxidants is uncertain, possibly within 20–35 % due to halons, about 25 % due to O(¹D), the rest being due to stratospheric OH (Neef et al., 2010). The oxidation of methane in the stratosphere produces significant amounts of water vapour, which has a positive radiative forcing, and stimulates the production of OH through its reaction with atomic oxygen (Forster et al., 2007). Stratospheric methane thus contributes significantly to the observed variability and trend in stratospheric water vapour (Hegglin et al., 2014). Uncertainties in the chemical loss of stratospheric methane are large, due to uncertain interannual variability in stratospheric transport as well as through its chemical interactions with stratospheric ozone (Portmann et al., 2012).

Unsaturated oxic soils are sinks of atmospheric methane due to the presence of methanotrophic bacteria, which consume methane as a source of energy. Wetlands with temporally variable saturation can also act as methane sinks. Dutaur and Verchot (2007) conducted a comprehensive meta-analysis of field measurements of CH₄ uptake spanning a variety of ecosystems. They reported a range of 36 ± 23 Tg CH₄ yr^-1 but also showed that stratifying the results by climatic zone, ecosystem and soil type led to a narrower range (and lower mean estimate) of 22 ± 12 Tg CH₄ yr^-1. A modelling study by Ridgwell et al. (1999) simulated the sink to be 20– 51 Tg CH₄ yr^-1. Curry (2007) used a process-based methane consumption scheme coupled to a land surface model (and calibrated to field measurements) to obtain a global estimate of 28 Tg CH₄ yr^-1, with a range of 9–47 Tg CH₄ yr^-1, which is the result reported in Kirschke et al. (2013). Tian et al. (2016) further updated the CH₄ uptake from soil, with the estimate of 30 ± 19 Tg CH₄ yr^-1. In that model, CH₄ uptake was determined by the diffusion rate of methane and oxygen through the uppermost soil layer, which was in turn dependent upon the soil characteristics (e.g. texture, bulk density) and water content (Curry, 2007). Riley et al. (2011) used another process-based model and estimated a global atmospheric CH₄ sink of 31 Tg CH₄ yr^-1. The methane consumption rate was also dependent on the available soil water, soil temperature and nutrient availability. Although not addressed in that model, it should be noted that if the soil water content increases enough to inhibit the diffusion of oxygen, the soil could become a methane source (Lohila et al., 2016). This transition can be rapid, thus creating areas that can be either a source or a sink of methane depending on the season.

Following Curry (2007), and consistent with Tian et al. (2015), we report here a climatological range of 9–47 Tg CH₄ yr^-1 as in Kirschke et al. (2013).

The global atmospheric lifetime is defined for a gas in steady state as the global atmospheric burden (Tg) of this gas divided by its global total sink (Tg yr^-1) (IPCC, 2001). In a case of a gas whose local lifetime is constant in space and time, the atmospheric lifetime equals the decay time (efolding time) of a perturbation. As methane is not in a steady state, we need to fit with a function that approaches steady state when calculating methane lifetime using atmospheric measurements (Sect. 4.1.1). Global models provide an estimate of the loss of the gas due to individual sinks, which can then be used to derive lifetime due to a specific sink. For example, methane’s tropospheric lifetime is determined as global atmospheric methane burden divided by the loss from OH oxidation in the troposphere, sometimes called “chemical lifetime”, while its total lifetime corresponds to the global burden divided by the total loss including tropospheric loss from OH oxidation, stratospheric chemistry and soil uptake. Recent multimodel estimate of the tropospheric methane lifetime is of 9.3 years (range 7.1–10.6; Voulgarakis et al., 2013; Kirschke et al., 2013) and that of the total methane lifetime is 8.2 ± 0.8 years (for year 2000, range 6.4–9.2, Voulgarakis et al., 2013). The model results for total methane lifetime are consistent with, though smaller than, the value reported in Table 6.8 of the IPCC AR5 of 9.1 ± 0.9 years (which was the observationally constrained estimate of Prather et al., 2012) most commonly used in the literature (Ciais et al., 2013) and the steady-state calculation from atmospheric observations (9.3 years, Sect. 4.1.1).

The first systematic atmospheric CH₄ observations began in 1978 (Blake et al., 1982) with infrequent measurements from discrete air samples collected in the Pacific at a range of latitudes from 67◦ N to 53◦ S. Because most of these air samples were from well-mixed oceanic air masses and the measurement technique was precise and accurate, they were sufficient to establish an increasing trend and the first indication of the latitudinal gradient of methane. Spatial and temporal coverage was greatly improved soon after (Blake and Rowland, 1986) with the addition of the NOAA flask network (Steele et al., 1987; Fig. 1), and of AGAGE (Cunnold et al., 2002), CSIRO (Francey et al., 1999), and other networks (e.g. ICOS network in Europe, https://www.icos-ri.eu/). The combined datasets provide the longest time series of globally averaged CH₄ abundance. Since the early 2000s, remotely sensed retrievals of CH₄ have provided CH₄ atmospheric columnaveraged mole fractions (Buchwitz et al., 2005a; Frankenberg et al., 2005; Butz et al., 2011; Crevoisier et al., 2009; Wunch et al., 2011). Fourier transform infrared (FTIR) measurements at fixed locations also provide methane column observations (Wunch et al., 2011).

Four observational networks provide globally averaged CH₄ mole fractions at the Earth’s surface: the Earth System Research Laboratory from US National Oceanic and Atmospheric Administration (NOAA/ESRL, Dlugokencky et al., 1994), the Advanced Global Atmospheric Gases Experiment (AGAGE, Prinn et al., 2000; Cunnold et al., 2002; Rigby et al., 2008), the Commonwealth Scientific and Industrial Research Organisation (CSIRO, Francey et al., 1999) and the University of California Irvine (UCI, Simpson et al., 2012). The data are archived at the World Data Centre for Greenhouse Gases (WDCGG) of the WMO Global Atmospheric Watch (WMO-GAW) programme, including measurements from other sites that are not operated as part of the four networks.

The networks differ in their sampling strategies, including the frequency of observations, spatial distribution, and methods of calculating globally averaged CH₄ mole fractions. Details are given in the Supplement of Kirschke et al. (2013). For the global average values of CH₄ concentrations presented here, all measurements are made using gas chromatography with flame ionization detection (GC/FID), although chromatographic schemes vary among the labs. Because GC/FID is a relative measurement method, the instrument response must be calibrated against standards. NOAA maintains the WMO CH₄ mole fraction scale X2004A; NOAA and CSIRO global means are on this scale. AGAGE uses an independent standard scale (Aoki et al., 1992), but direct comparisons of standards and indirect comparisons of atmospheric measurements show that differences are below 5 ppb (WMO RoundRobin programme). UCI uses another independent scale that was established in 1978 and is traceable to NIST (Simpson et al., 2012) but has not been included in standard exchanges with other networks so differences with the other networks cannot be quantitatively defined. Additional experimental details are presented in the Supplement from Kirschke et al. (2013) and references therein.

In Fig. 1, (a) globally averaged CH₄ and (b) its growth rate (derivative of the deseasonalized trend curve) through 2012 are plotted for a combination of the four measurement programmes using a procedure of signal decomposition described in Thoning et al. (1989). We define the annual increase GATMas the increase in the growth rate from 1 January in one year to 1 January in the next year. Agreement among the four networks is good for the global growth rate, especially since ∼ 1990. The long-term behaviour of globally averaged atmospheric CH₄ shows a decreasing but positive growth rate (defined as the derivative of the deseasonalized mixing ratio) from the early 1980s through 1998, a near-stabilization of CH₄ concentrations from 1999 to 2006, and a renewed period with positive but stable growth rates since 2007. When a constant atmospheric lifetime is assumed, the decreasing growth rate from 1983 through 2006 implies that atmospheric CH₄ was approaching steady state, with no trend in emissions. The NOAA global mean CH₄ concentration was fitted with a function that describes the approach to a first-order steady state (SSindex): [CH₄](t ) = [CH₄]SS − ([CH₄]SS − [CH₄]0)e−t/τ; solving for the lifetime, τ , gives 9.3 years, which is very close to current literature values (e.g. Prather et al., 2012).

On decadal timescales, the annual increase is on average 2.1 ± 0.3 ppb yr^-1 for 2000–2009, 3.5 ± 0.2 ppb yr^-1 for 2003–2012 and 5.0 ± 1.0 ppb yr^-1 for the year 2012. The two decadal values hide a jump in the growth rate after 2006. Indeed, from 1999 to 2006, the annual increase of atmospheric CH₄ was remarkably small at 0.6 ± 0.1 ppb yr^-1. In the last 8 years, the atmospheric growth rate has recovered to a level similar to that of the mid-1990s (∼ 5 ppb yr^-1), before the stabilization period of 1999–2006, as stated in Kirschke et al. (2013).

In the 2000s, two space-borne instruments sensitive to atmospheric methane were put in orbit and have provided atmospheric methane column-averaged dry air mole fraction (XCH₄), using either shortwave infrared spectrometry (SWIR) or thermal infrared spectrometry (TIR).

Between 2003 and 2012, the Scanning Imaging Absorption spectrometer for Atmospheric CartograpHY (SCIAMACHY) was operated on board the ESA ENVIronmental SATellite (ENVISAT), providing nearly 10 years of XCH₄ sensitive to the atmospheric boundary layer (Burrows et al., 1995; Buchwitz et al., 2006; Dils et al., 2006; Frankenberg et al., 2011). These satellite retrievals were the first to be used for global and regional inverse modelling of methane fluxes (Meirink et al., 2008a; Bergamaschi et al., 2007, 2009). The relatively long time record allowed the analysis of the interannual methane variability (Bergamaschi et al., 2013). However, the use of SCIAMACHY necessitates important bias correction, especially after 2005 (up to 40 ppb from south to north) (Bergamaschi et al., 2009; Houweling et al., 2014; Alexe et al., 2015).

In January 2009, the JAXA satellite Greenhouse Gases Observing SATellite (GOSAT) was launched containing the TANSO-FTS instrument, which observes in the shortwave infrared (SWIR). Different retrievals of methane based on TANSO-FTS/GOSAT products are made available to the community (Yoshida et al., 2013; Schepers et al., 2012; Parker et al., 2011) based on two retrieval approaches: proxy and full physics. The proxy method retrieves the ratio of methane column (XCH₄) and carbon dioxide column (XCO₂), from which XCH₄ is derived after multiplication with transport model-derived XCO2 (Chevallier et al., 2010; Peters et al., 2007; Frankenberg et al., 2006). It intends mostly to remove biases due to light scattering on clouds and aerosols and is highly efficient owing to the small spectral distance between CO2 and CH₄ sunlight absorption bands (1.65 µm for CH₄ and 1.60 µm for CO₂). Because of this, scattering-induced errors are similar for XCO2 and XCH₄ and cancel out in the ratio. The second approach is the full-physics algorithm, which retrieves the aerosol properties (amount, size and height) along with CO2 and CH₄ columns (e.g. Butz et al., 2011). Although GOSAT retrievals still show significant unexplained biases (possibly also linked to atmospheric transport modelling; Locatelli et al., 2015) and limited sampling in cloud-covered regions and in the high-latitude winter, it represents an important improvement compared to SCIAMACHY both for random and systematic observation errors (see Table S2 of Buchwitz et al., 2016).

Atmospheric inversions based on SCIAMACHY or GOSAT CH₄ retrievals have been carried out by different research groups (Monteil et al., 2013; Cressot et al., 2014; Alexe et al., 2015; Bergamaschi et al., 2013; Locatelli et al., 2015). For GOSAT, differences between the use of proxy and full-physics retrievals have been investigated. In addition, joint CO₂–CH₄ inversions have been conducted to investigate the use of GOSAT retrieved ratios avoiding a modelderived hard constraint on XCO₂ (Pandey et al., 2015, 2016; Fraser et al., 2013). Results from some of these studies are reported in Sect. 5 of this paper.

The processes emitting methane discriminate differently its isotopologues (isotopes). The two main stable isotopes of CH₄ are 13CH₄ and CH3D,and there is also the radioactive carbon isotope 14C–CH₄. Isotopic signatures are conventionally given by the deviation of the sample mole ratio (for example, R=13CH₄/12CH₄or CH3D/ CH₄) relative to a given standard (Rstd)relative to a reference ratio, given in per mil as in Eq. (3).

δ13CH₄ or δD (CH₄) >= − 1 × 1000 (3)Rstd

For the 13CH₄ isotope, the conventional reference standard is known as Vienna Pee Dee Belemnite (VPDB), with Rpdb = 0.0112372. The same definition applies to CH3D,with the Vienna Standard Mean Ocean Water (VSMOW) RSMOW= 0.00015575. The isotopic composition of atmospheric methane is measured at a subset of surface stations (Quay et al., 1991, 1999; Lowe et al., 1994; Miller et al., 2002; Morimoto et al., 2006; Tyler et al., 2007). The mean atmospheric values are about −47 ‰ for δ13CH₄ and −86/−96 ‰ for δD(CH₄). Isotopic measurements are made mainly on flask air samples analysed with gaschromatograph isotope ratio spectrometry for which an accuracy of 0.05 ‰ for δ13CH₄ and 1.5 ‰ for δD(CH₄) can be achieved (Rice et al., 2001; Miller et al., 2002). These isotopic measurements based on air flask sampling have relatively low spatial and temporal resolutions. Laser-based absorption spectrometers and isotope ratio mass spectrometry techniques have recently been developed to increase sampling frequency and allow in situ operation (McManus et al., 2010; Santoni et al., 2012).

Measurements of δ13CH₄ can help to partition the different methanogenic processes of methane: biogenic (−70 to −55 ‰), thermogenic (−55 to −25 ‰) or pyrogenic (−25 to −15 ‰) sources (Quay et al., 1991; Miller et al., 2002; Fisher et al., 2011) or even the methanogenic pathway (McCalley et al., 2014). δD(CH₄) provides valuable information on the oxidation by the OH radicals (Röckmann et al., 2011) due to a fractionation of about 300 ‰. Emissions also show substantial differences in δD(CH₄) isotopic signatures: −200 ‰ for biomass burning sources vs. −360 to −250 ‰ for biogenic sources (Melton et al., 2012; Quay et al., 1999). 14C– CH₄ measurements (Quay et al., 1991, 1999; Lowe et al., 1988) may also help to partition for fossil fuel contribution (radiocarbon-free source). For example, Lassey et al. (2007a) used more than 200 measurements of radioactive 14C–CH₄ (with a balanced weight between Northern and Southern hemispheres) to further constrain the fossil fuel contribution to the global methane source emission to 30 ± 2 % for the period 1986–2000.

Integrating isotopic information is important to improve our understanding of the methane budget. Some studies have simulated such isotopic observations (Neef et al., 2010; Monteil et al., 2011) or used them as additional constraints to inverse systems (Mikaloff Fletcher et al., 2004; Hein et al., 1997; Bousquet et al., 2006; Neef et al., 2010; Thompson et al., 2015). Using pseudo-observations, Rigby et al. (2012) found that quantum-cascade-laser-based isotopic observations would reduce the uncertainty in four major source categories by about 10 % at the global scale (microbial, biomass burning, landfill and fossil fuel) and by up to 50 % at the local scale. Although all source types cannot be separated using 13C, D and 14C isotopes, such data bring valuable information to constrain groups of sources in atmospheric inversions, if the isotopic signatures of the various sources can be precisely assessed (Bousquet et al., 2006, Supplement).

Other types of methane measurements are available, which are not commonly used to infer fluxes from inverse modelling (yet) but are used to verify its performance (see e.g. Bergamaschi et al., 2013). Aircraft or balloon-borne in situ measurements can deliver vertical profiles with high verti cal resolution. Such observations can also be used to test remote-sensing measurement from space or from the surface and bring them on the same scale as the in situ surface mea surements. Aircraft measurements have been undertaken in various regions either during campaigns (Wofsy, 2011; Beck et al., 2012; Chang et al., 2014; Paris et al., 2010) or in a recurrent mode using small aircrafts in the planetary bound ary layer (Sweeney et al., 2015; Umezawa et al., 2014; Gatti et al., 2014) and commercial aircrafts (Schuck et al., 2012; Brenninkmeijer et al., 2007; Umezawa et al., 2012, 2014; Machida et al., 2008). Balloons can carry in situ instruments (e.g. Joly et al., 2008; using tunable laser diode spectrome try) or air samplers (e.g. air cores, Karion et al., 2010) up to 30 km height. New technologies have also developed systems based on cavity ring-down spectroscopy (CRDS), opening a large ensemble of new activities to estimate methane emissions such as drone measurements (light version of CRDS), as land-based vehicles for real-time, mobile monitoring over oil and gas facilities, as well as ponds, landfills, livestock, etc.

In October 2006, the Infrared Atmospheric Sounding In terferometer (IASI) on board the European MetOp-A satel lite began to operate. Measuring the thermal radiation from Earth and the atmosphere in the TIR, it provides mid-to upper troposphere columns of methane (representative of the 5–15 km layer) over the tropics using an infrared sounding interferometer (Crevoisier et al., 2009). Despite its sensitivity being limited to the mid-to-upper troposphere, its use in flux inversions has shown consistent results in the tropics with surface and other satellite-based inversions (Cressot et al., 2014).

The Total Carbon Column Observing Network (TCCON) uses ground-based Fourier transform spectrometers to measure atmospheric column abundances of CO²,, CO, CH₄, N2O and other molecules that absorb sunlight in the near infrared spectral region (Wunch et al., 2011). As TCCON measurements make use of sunlight, they can be performed throughout the day during clear-sky conditions, with the sun typically 10◦ above the horizon. The TCCON network has been established as a reference for the validation of column retrievals, like those from SCIAMACHY and GOSAT. TC CON data can be obtained from the TCCON Data Archive, hosted by the Carbon Dioxide Information Analysis Center (CDIAC, http://cdiac.ornl.gov/).

An atmospheric inversion for methane fluxes (sources and sinks) optimally combines atmospheric observations of methane and associated uncertainties, a prior knowledge of the fluxes including their uncertainties, and a chemistry transport model to relate fluxes to concentrations (Rodgers, 2000). In this sense, top-down inversions integrate all the components of the methane cycle described previously in this paper. The observations can be surface or upper-air in situ observations, as well as satellite and surface retrievals. Prior emissions generally come from bottom-up approaches such as process-based models or data-driven extrapolations (nat ural sources) and inventories (anthropogenic sources). The chemistry transport model can be Eulerian or Lagrangian, and global or regional, depending on the scale of the flux to be optimized. Atmospheric inversions generally rely on the Bayes’ theorem, which leads to the minimization of a cost function as Eq. (4): (See original document)

where y is a vector containing the atmospheric observations, x is a state vector containing the methane emissions and other appropriate variables (like OH concentrations or CH₄ concentrations at the start of the assimilation window) to be estimated, xbis the prior state of x, and H is the observation operator, here the combination of an atmospheric transport and chemistry model and an interpolation procedure sam pling the model at the measurement coordinates. R is the er ror covariance matrix of the observations and Pb is the error covariance matrix associated with xb. The errors on the mod elling of atmospheric transport and chemistry are included in the R matrix (Tarantola, 1987). The minimization of a lin earized version of J leads to the optimized state vector xa(Eq. 5): (See original document) where Pa is given by Eq. (6) and represents the error covari ance matrix associated with xa, and H contains the sensitiv ities of any observation to any component of state vector x (linearized version of the observation operator H (x)).(See original document)

A group of eight atmospheric inversion systems using global Eulerian transport models were used in this synthesis. Each inversion system provides from 1 to 10 inversions, including sensitivity tests varying the assimilated observations (surface or satellite) or the inversion setup. This represents a total of 30 inversion runs with different time coverage: generally 2000–2012 for surface-based observations, 2003–2012 for SCIAMACHY-based inversions and 2009–2012 for GOSATbased inversions (Table 3). When multiple sensitivity tests were performed we use the mean of this ensemble not to overweight one particular inverse model. Bias correction procedures have been developed to assimilate SCIAMACHY (Bergamaschi et al., 2009, 2013; Houweling et al., 2014) and GOSAT data (Cressot et al., 2014; Houweling et al., 2014; Locatelli et al., 2015; Alexe et al., 2015). These procedures can lead to corrections from several parts per billion and up to several tens of parts per billion (Bergamaschi et al., 2009; Locatelli et al., 2015). Although partly due to transport model errors, the large corrections applied to satellite total column CH₄ data question the comparably low systematic errors reported in satellite validation studies using TCCON (Dils et al., 2014; CCI-Report, 2016). It should also be noticed that some satellite-based inversions are in fact combined satellite and surface inversions as they use either instantaneous in situ data simultaneously (Bergamaschi et al., 2013; Alexe et al., 2015) or annual mean surface observations to correct satellite bias (Locatelli et al., 2015). Nevertheless, these inversions are still referred to as satellite-based inversions.

General characteristics of the inversion systems are provided in Table 3. Further detail can be found in the referenced papers. Each group was asked to provide gridded flux estimates for the period 2000–2012, using either surface or satellite data, but no additional constraints were imposed so that each group could use their preferred inversion setup. This approach is appropriate for our purpose of flux assessment but not necessarily for model intercomparison. We did not require posterior uncertainty from the different participating groups, which may be done for the next release of the budget. Indeed chemistry transport models have some limitations that impact on the inferred methane budget, such as discrepancies in interhemispheric transport, stratospheric methane profiles and OH distribution. We consider here an ensemble of inversions gathering a large range of chemistry transport models, through their differences in vertical and horizontal resolutions, meteorological forcings, advection and convection schemes and boundary layer mixing; we assume that this model range is sufficient to cover the range of transport model errors in the estimate of methane fluxes. Each group provided gridded monthly maps of emissions for both their prior and posterior total and for sources per category (see the categories Sect. 2.3). Results are reported in Sect. 5. Atmospheric sinks were not analysed for this budget, which still relies on Kirschke et al. (2013) for bottom-up budget and on a global mass balance for top-down budget (difference between the global source and the observed atmospheric increase).

The last year of reported inversion results is 2012, which represents a 4-year lag with the present. Satellite observations are linked to operational data chains and are generally available within days to weeks after the recording of the spectra. Surface observations can lag from months to years because of the time for flask analyses and data checks in (mostly) non-operational chains. With operational networks such as ICOS in Europe, these lags will be reduced in the future. In addition, the final 6 months of inversions are generally ignored (spun down) because the estimated fluxes are not constrained by as many observations as the previous months. Finally, the long inversion runs and analyses can take up to months to be performed. For the next global methane budget the objective is to represent more recent years by reducing the analysis time and shortening the in situ atmospheric observation release.

At the global scale, the total emissions inferred by the ensemble of 30 inversions are 558 Tg CH₄ yr^-1 [540–570] for the 2003–2012 decade (Table 4), with a higher value of 568 Tg CH₄ yr^-1 [542–582] for 2012. Global emissions for 2000–2009 (552 Tg CH₄ yr^-1) are consistent with Kirschke et al. (2013), and the range of uncertainties for global emissions (535–566) is in line as well with that of Kirschke et al. (2013) (526–569), although 8 out of the 30 inversions presented here (∼ 25 %) are different. The latitudinal breakdown of emissions inferred from atmospheric inversions reveals a dominance of tropical emissions at 359 Tg CH₄ yr^-1 [339– 386], representing 64 % of the global total. Thirty-two per cent of the emissions are from the midlatitudes and 4 % from high latitudes (above 60◦ N).

The picture given by the bottom-up approaches is quite different with global emissions of 736 Tg CH₄ yr^-1 [596–884] for 2003–2012 (Table 2). This estimate is much larger than top-down estimates. The bottom-up estimate is given by the sum of individual anthropogenic and natural processes, with no constraint on the total. As noticed in Kirschke et al. (2013), such a large global emissions rate is not consistent with atmospheric constraints brought by OH optimization and is very likely overestimated. This overestimation likely results from errors in the estimation of natural sources and sinks: extrapolation or double counting of some natural sources (e.g. wetlands, inland waters), or estimation of atmospheric sink terms. The anthropogenic sources are much more consistent between bottom-up and top-down approaches (Sect. 5.1.2).

The global methane budget for five source categories (see Sect. 2.3) for 2003–2012 is presented in Fig. 5 and Table 2. Top-down estimates attribute about 60 % of the total emissions to anthropogenic activities (range of 50–70 %) and 40 % to natural emissions. As natural emissions from bottom-up models are much larger, the anthropogenic vs. natural emission ratio is more balanced for bottom-up (∼ 50 % each). A predominant role of anthropogenic sources of methane emissions is strongly supported by the ice core and atmospheric methane records. The data indicate that atmospheric methane varied around 700 ppb during the last millennium before increasing by a factor of 2.6 to ∼ 1800 ppb. Accounting for the decrease in mean lifetime over the industrial period, Prather et al. (2012) estimate from these data a total source of 554 ± 56 Tg CH₄ in 2010 of which about 64 % (352 ± 45 Tg CH₄) are of anthropogenic origin, very consistent estimates with our synthesis.

For 2003–2012, the top-down and bottom-up derived estimates of respectively 167 Tg CH₄ yr^-1 (range 127–202) and 185 Tg CH₄ yr^-1 (range 153–227) are statistically consistent. Mean wetland emissions for the 2000–2009 period appear similar, albeit slightly smaller than found in Kirschke et al. (2013): 166 Tg CH₄ yr−1 in this study vs. 175 Tg CH₄ yr^-1 in Kirschke et al. (2013) for top-down (−4 %) and 183 Tg CH₄ yr^-1 in this study vs. 217 Tg yr^-1 in Kirschke et al. (2013) for bottom-up (−15 %). Note that more inversions (top-down) and more wetland models

(bottom-up) were used in this study. Inversions have difficulty in separating wetlands from other sources so that uncertainties on top-down wetland emissions remain large. In this study, all bottom-up models were forced with the same wetland extent and climate forcings (Poulter et al., 2016), with the result that the amplitude of the range of emissions of 151–222 for 2000–2009 has narrowed by a third compared to the previous estimates from Melton et al. (2013) (141–264) and from Kirschke et al. (2013) (177–284). This suggests that differences in wetland extent explain about a third (30–40 %) of the former range of the emission estimates of global natural wetlands. The remaining range is due to differences in model structures and parameters. It is also worth noting that bottom-up and top-down estimates differ less in this study (∼ 17 Tg yr^-1 for the mean) than in Kirschke et al. (2013) (∼ 30 Tg yr^-1), although results from many more models are reported here. For top-down inversions, natural wetlands represent 30 % on average of the total methane emissions but only 25 % for bottom-up models (because of higher total emissions inferred by bottom-up models).

The discrepancy between top-down and bottom-up budgets is the largest for the natural emission total, which is 384 Tg CH₄ yr^-1 [257–524] for bottom-up and only 231 Tg CH₄ yr^-1 [194–296] for top-down over the 2003– 2012 decade. Processes other than natural wetlands (Fig. 5), namely freshwater systems, geological sources, termites, oceans, wild animals, wildfires, and permafrost, explain this large discrepancy. For the 2003–2012 decade, topdown inversions infer non-wetland natural emissions of 64 Tg CH₄ yr^-1 [21–132], whereas the sum of the individual bottom-up emissions is 199 Tg CH₄ yr^-1 [104–297]. The two main contributors to this large bottom-up total are freshwater (∼ 60 %) and geological emissions (∼ 20 %), both of which have large uncertainties without spatially explicit representation. Because of the discrepancy, this category represents 10 % of total emissions for top-down inversions but 27 % for bottom-up approaches.

Improved area estimates of freshwater emissions would be beneficial. For example, stream fluxes are difficult to assess because of the high-expected spatial variability and very uncertain areas of headwater streams where methane-rich groundwater may be rapidly degassed. There are also uncertainties in the geographical distinction between wetlands, small lakes (e.g. thermokarst lakes), and floodplains that will need more attention to avoid double counting. In addition, major uncertainty is still associated with representation of ebullition. The intrinsic nature of this large but very locally distributed flux highlights the need for cost-efficient highresolution techniques for resolving the spatio-temporal variations of these fluxes. In this context of observational gaps in space and time, freshwater fluxes are considered underestimated until measurement techniques designed to properly account for ebullition become more common (Wik et al., 2016a). On the contrary, global estimates for freshwater emissions rely on upscaling of uncertain emission factors and emitting areas, with probable overlapping of wetland emissions (Kirschke et al., 2013), which may also lead to an overestimate. More work is needed, based on both observations and process modelling, to overcome these uncertainties.

For geological emissions, relatively large uncertainties come from the extrapolation of only a subset of direct measurements to estimate the global fluxes. Moreover, marine seepage emissions are still widely debated (Berchet et al., 2016), and particularly diffuse emissions from microseepage are highly uncertain. However, summing up all fossilCH₄-related sources (including the anthropogenic emissions) leads to a total of 173 Tg CH₄ yr−1 [149–209], which is about 31 % [25–35 %] of global methane emissions. This result is consistent with ¹4C atmospheric isotopic analyses inferring a 30 % contribution of fossil-CH₄ to global emissions (Lassey et al., 2007b; Etiope et al., 2008). All nongeological and non-wetland land source categories (wild animals, wildfires, termites, permafrost) have been evaluated at a lower level than in Kirschke et al. (2013) and contribute only 23 Tg CH₄ yr^-1 [9–36] to global emissions. From a topdown point of view, the sum of all natural sources is more robust than the partitioning between wetlands and other natural sources. To reconcile top-down inversions and bottom up estimates, the estimation and proper partition of methane emissions from wetlands and freshwater systems should receive high priority.

Total anthropogenic emissions are found statistically consistent between top-down (328 Tg CH₄ yr^-1, range 259–370) and bottom-up approaches (352 Tg CH₄ yr^-1, range 340– 360), although top-down average is about 7 % smaller than bottom-up average over 2003–2012. The partition of anthropogenic emissions between agriculture and waste, fossil fuel extraction and use, and biomass and biofuel burning also shows good consistency between top-down and bottom-up approaches (Table 2 and Fig. 7). For 2003–2012, agriculture and waste contributed 188 Tg CH₄ yr^-1 [115–243] for top-down and 195 Tg CH₄ yr^-1 [178–206] for bottom-up. Fossil fuel emissions contributed 105 Tg CH₄ yr^-1 [77–133] for top-down and 121 Tg CH₄ yr^-1 [114–133] for bottomup. Biomass and biofuel burning contributed 34 Tg CH₄ yr^-1 [15–53] for top-down and 30 Tg CH₄ yr^-1 [27–35] for bottom-up. Biofuel methane emissions rely on very few estimates at the moment (Wuebbles and Hayhoe, 2002; GAINS model). Although biofuel is a small source globally (∼ 12 Tg CH₄ yr^-1), more estimates are needed to allow a proper uncertainty assessment. Overall for top-down inversions the global fraction of total emissions for the different source categories are 33 % for agriculture and waste, 20 % for fossil fuels, and 6 % for biomass and biofuel burnings. With the exception of biofuel emissions, the global uncertainty of anthropogenic emissions appears to be smaller than that of natural sources but with asymmetric uncertainty distribution (mean significantly different than median). In poorly observed regions, top-down inversions rely on the prior estimates and bring little or no additional information to constrain the (often) spatially overlapping emissions (e.g. in India, China). Therefore, the relative agreement between top-down and bottom-up may indicate the limited capability of the inversion to separate the emissions and should therefore be treated with caution. Although the uncertainty range of some emissions has been decreased in this study compared to Kirschke et al. (2013) (e.g. oceans, termites, geological), there is no uncertainty reduction in the regional budgets because of the larger range reported for emissions from freshwater systems.

At regional scale, for the 2003–2012 decade (Table 4 and Fig. 6), total methane emissions are dominated by Africa with 86 Tg CH₄ yr^-1 [73–108], tropical South America with a total of 84 Tg CH₄ yr^-1 [65–101], and South East Asia with 73 Tg CH₄ yr^-1 [55–84]. These three (mainly) tropical regions represent almost 50 % of methane emissions worldwide. The other high-emitting source regions are China (58 Tg CH₄ yr^-1 [51-72]), central Eurasia and Japan (46 Tg CH₄ yr^-1 [38–54]), contiguous USA (41 Tg CH₄ yr^-1 [34–49]), Russia (38 Tg CH₄ yr^-1 [31–44]), India (39 Tg CH₄ yr^-1 [37–46]) and Europe (28 Tg CH₄ yr^-1 [21–34]). The other regions (boreal and central North America, temperate South America, Oceania, oceans) contribute between 7 and 20 Tg CH₄ yr^-1. This budget is consistent with Kirschke et al. (2013) within the large ranges around the mean emissions, although larger emissions are found here for South America, South East Asia, and Europe and lower emissions are found for Africa, North America and China. The regions with the largest changes are usually the least constrained by the surface networks.

The different inversions assimilated either satellite or ground-based observations. It is of interest to determine whether these two types of data provide consistent surface emissions. To do so, we computed global, hemispheric and regional methane emissions using satellite-based inversions and ground-based inversions separately for the 2010–2012 time period, which is the longest time period for which results from both GOSAT satellite-based and surface-based inversions were available. At the global scale, satellite-based inversions infer significantly higher emissions (>+12 Tg CH₄ yr^-1, p >= 0.04) than ground-based in versions. At the regional scale, emissions varied between the satellite-based and surface-based inversions, although the difference is not statistically significant due to too few inversions and some outliers making the ensemble not robust enough. Yet the largest differences (satellite-based minus surface based inversions) are observed over the tropical region: tropical South America >+11 Tg CH₄ yr^-1; southern Africa >+6 Tg CH₄ yr^-1; India −6 Tg CH₄ yr^-1; and over China −7 Tg CH₄ yr^-1. Satellite data provide more constraints on fluxes in tropical regions than surface-based inversions, due to a much larger spatial coverage. It is therefore not surprising that most differences between these two types of observations are found in the tropical band. However, such differences could also be due to the larger systematic errors of satellite data as compared to surface networks (Dils et al., 2014). In this context, the way the stratosphere is treated in the atmospheric models used to produce atmospheric methane columns from remote-sensing measurements (e.g. GOSAT or TCCON) seems important to further investigate (Locatelli et al., 2015; Monteil et al., 2011; Bergamaschi et al., 2009). Recent papers have developed methodologies to extract tropospheric partial column abundances from the TCCON data (Saad et al., 2014; Wang et al., 2014). Such partitioning could help explain the discrepancies between atmospheric models and satellite data.

The analysis of the regional methane budget per source category (Fig. 7) can be performed both for bottom-up and top-down approaches but with limitations. A complementary view of the methane budget is also available as an interactive graphic produced using data visualization techniques (http://lsce-datavisgroup.github.io/MethaneBudget/). Moving the mouse over regions, processes or fluxes reveals their relative weights in the global methane budget and provides the mean values and the minimum–maximum ranges of their contributions (mean [min, max]). The total source estimates from the bottom-up approaches are further classed into finer subcategories. This graphic shows that there is good consistency between top-down and bottom-up approaches in the partition of anthropogenic emissions between agriculture and waste, fossil fuel extraction and use, and biomass and biofuel burning, and it also highlights the disequilibrium between top-down (left) and bottom-up (right) budgets, mainly due to natural sources. On the bottom-up side, some natural emissions are not (yet) available at regional scale (oceans, geological, inland waters). Therefore, the category “others” is not shown for bottom-up results in Fig. 7 and is not regionally attributed in the interactive graphic. On the top-down side, as already noted, the partition of emissions per source category has to be considered with caution. Indeed, using only atmospheric methane observations to constrain methane emissions makes this partition largely dependent on prior emissions. However, differences in spatial patterns and seasonality of emissions can still be constrained by atmospheric methane observations for those inversions solving for different sources categories (see Sect. 2.3).

Wetland emissions largely dominate methane emissions in tropical South America, boreal North America, southern Africa, temperate South America and South East Asia, although agriculture and waste emissions are almost as important for the last two regions. Agriculture and waste emissions dominate in India, China, contiguous USA, central North America, Europe and northern Africa. Fossil fuel emissions dominate in Russia and are close to agriculture and waste emissions in the region called central Eurasia and Japan. In China, fossil fuel emissions are on average close, albeit smaller, than agriculture and waste emissions. Comparison between bottom-up and top-down approaches shows good consistency, but one has to consider the generally large error bars, especially for top-down inversions. The largest discrepancy occurs for wetland emissions in boreal North America where bottom-up models infer larger emissions (32 Tg CH₄ yr^-1) than top-down inversions (13 Tg CH₄ yr^-1). Indeed, one particular bottom-up model infers a 61 Tg CH₄ yr^-1 emission for this region, largely above estimates from other models, which lie between 15 and 45 Tg CH₄ yr^-1. Top-down models results are consistent with the climatology proposed by Kaplan (2002), whereas bottom-up models are more in line, albeit larger, than the climatology of Matthews and Fung (1987), who infer about 30 Tg CH₄ yr^-1 for boreal North America. Interestingly, the situation is different for Russia where top-down and bottomup approaches show similar mean emissions from natural wetlands (mostly boreal, ∼ 13–14 Tg CH₄ yr^-1), consistently with Kaplan (2002) but not with Matthews and Fung (1987), who infer almost 50 Tg CH₄ yr−1 for Russia.

Wetland emissions from Russia appear very uncertain, as also found by Bohn et al. (2015) for western Siberia. Wetland emissions from tropical South America are found more consistent in this work than in Kirschke et al. (2013), where topdown inversions showed 2 times less emission than bottomup models. The larger number of bottom-up models (11 against 3) and top-down inversions (30 against 8) are plausible causes explaining the improved agreement in this tropical region, poorly constrained by the surface networks (Pison et al., 2013).

Anthropogenic emissions remain close between top-down and bottom-up approaches for most regions, again with the possibility that part of this agreement is due to the lack of information brought by atmospheric observations to top-down inversions for some regions. One noticeable exception is the lower emissions for China as compared to the prior, visible also in Fig. 6. A priori anthropogenic emissions for China are mostly provided by the EDGARv4.2 inventory. Starting from prior emissions of 67 Tg CH₄ yr^-1 [58–77], the mean of the atmospheric derived estimates for China is 58 Tg CH₄ yr^-1 [51–72], corresponding to a −14 % difference of the Chinese emissions. A t test performed for the available estimates suggests that the mean posterior total emission for China is different from the prior emission at the 95 % confidence level. Several atmospheric studies have already suggested a possible overestimation of methane emissions from coal in China in the EDGARv4.2 inventory (Bergamaschi et al., 2013; Kirschke et al., 2013; Tohjima et al., 2014; Umezawa et al., 2014). Indeed, comparing the results of top-down inversions to EDGARv4.2 inventory (maximum of bottom-up estimates for China in Fig. 7), fossil fuel emissions are reduced by 33 % from 30 to 20 Tg CH₄ yr^-1 (range 9–30) and agriculture and waste emissions are reduced by 27 % from 37 to 27 Tg CH₄ yr^-1 (range 16–37). This result is consistent with a new inventory for methane emissions from China based on county-scale data (43 ± 6 Tg yr^-1), indicating that coal-related methane emissions are 37 % (−7 Tg yr^-1) lower than reported in the EDGARv4.2 inventory (Peng et al.,2016) (see also Sect. 3.1.2). Thompson et al. (2015) showed that their prior (based on EDGARv4.2) overestimated the Chinese methane emissions by 30 %; however, they found no significant difference in the coal sector estimates between prior and posterior and attribute the difference to rice emissions. It demonstrates that inversions are capable of verifying regional emissions when biases in the inventories are substantial, as in the case of China.

In contrast to the Chinese estimates, emissions inferred for Africa and especially southern Africa are significantly larger than in the prior estimates (Fig. 6). For example, for southern Africa, the mean of the inversion ensemble is 44 Tg CH₄ yr−1 [37–53], starting at a mean prior of 36 Tg CH₄ yr^-1 [27–35]. This is a 25 % increase compared to mean prior estimates for southern Africa. A t test performed for the available estimates suggests that the mean posterior for southern Africa is different from the prior at the 98 % confidence level. An increase of northern African emissions is also inferred from the ensemble of inversions but is less significant.

For all other regions, emission changes compared to prior values remain within the first and third quartiles of the distributions. In particular, contiguous USA (without Alaska) is found to emit 41 Tg CH₄ yr^-1 [34–49], which is close to the prior estimates. Top-down and bottom-up estimates are consistent for anthropogenic sources in this region. Only natural wetlands are lower as estimated by top-down models (9 Tg CH₄ yr^-1 [6–13]) than by bottom-up models (13 Tg CH₄ yr^-1 [6–23]).

Kirschke et al. (2013) identified four main shortcomings in the assessment of regional to global CH₄ budgets, which we revisit now.

Annual to decadal CH₄ emissions from natural sources (wetlands, fresh water, geological) are highly uncertain. The work by Poulter et al. (2016), following Melton et al. (2013) allows partitioning the uncertainty (expressed as the range in the estimates) of methane emissions from natural wetlands between wetland extent and other components, based on the use of a common and newly developed database for wetland extent. This approach confirms that wetland extent dominates the uncertainty of modelled methane emissions from wetlands (30–40 % of the uncertainty). The rest of the uncertainty lies in the model parameterizations of the flux density, which remains poorly constrained due to very few methane flux measurements available for different ecosystems over time. More measurements of the isotopic atmospheric composition of the various ecosystems (bogs/swamps, C3/C4 vegetation, etc.) would also help better constrain methane fluxes as well as its isotopic signature in the wetland models. In addition, the footprints of flux measurements are largely on too small scales (e.g. chamber measurements) to be compared with the lower resolution at which land surface models operate. Although more and more flux sites now integrate measurements of methane fluxes by eddy covariance, such a technique can reveal unexpected issues (e.g. Baldocchi et al., 2012). There is a need for integration of methane flux measurements on the model of the FLUXNET activity (http://fluxnet.ornl.gov/). This would allow further refinement of the model parameterizations (Turetsky et al., 2014; Glagolev et al., 2011). A comparison of the model ensemble estimates against bottom-up inventory for western Siberia by Glagolev et al. (2011) made by Bohn et al. (2015) showed that there still is a sizable disagreement between their results. A more complete analysis of the literature for freshwater emissions has led to a 50 % increase of the reported range compared to Kirschke et al. (2013). Emitting pathways such as ebullition remain poorly understood and quantified. There is a need for systematic measurements from a suite of sites reflecting the diversity of lake morphologies to better understand the short-term biological control on ebullition variability (Wik et al., 2014). Similarly more local measurements using continuous-laser-based techniques would allow refining the estimation of geological methane emissions. Further efforts are needed: (1) extending the monitoring of the methane emissions from the different natural sources (wetlands, fresh waters and geological) complemented with key environmental variables to allow proper interpretation (e.g. soil temperature and moisture, vegetation types, water temperature, acidity, nutrient concentrations, NPP, soil carbon density); (2) developing process-based modelling approaches to estimate inland emissions instead of data-driven extrapolations of unevenly distributed and local flux observations; and (3) creating a global flux product for all inland water emissions at high resolution allowing the avoidance of double counting between wetlands and freshwater systems.

The partitioning of CH₄ emissions and sinks by region and process is not sufficiently constrained by atmospheric observations in top-down models. In this work, we report inversions assimilating satellite data from GOSAT (and one inversion using SCIAMACHY), which bring more constraints, especially over tropical continents. The extension of the CH₄ surface networks to poorly observed regions (e.g. tropics, China, India, high latitudes) is still critical to complement satellite data, which do not observe well in cloudy regions and at high latitudes but also to evaluate and correct satellite biases. Such data now exist for China (Fang et al., 2015), India (Tiwari and Kumar, 2012; Lin et al., 2015) and Siberia (Sasakawa et al., 2010; Winderlich et al., 2010) and can be assimilated in inversions in the upcoming years. Observations from other tracers could help partition the different methane emitting processes. Carbon monoxide (FortemsCheiney et al., 2011) can provide constraints for biomass burning for instance. However, additional tracers can also bring contradictory trends in emissions such as the ones suggested since 2007 by ¹3C (Schaefer et al., 2016) and ethane (Hausmann et al., 2016). Such discrepancies have to be understood and solved to be able to properly use additional tracers to constrain methane emissions. An update of OH fields is expected in 2016 with an ensemble of chemistry transport model and chemistry-climate model simulations in the framework of CCMI (Chemistry-Climate Model Initiative) spanning the past 3 decades (http://www.met.reading.ac.uk/ ccmi/). The outcome of this experiment will contribute to an improved representation of the methane sink (Lamarque et al., 2013). The development of regional components of the global methane budget is also a way to improve global totals by developing regional top-down and bottom-up approaches. Such efforts are underway for South and East Asia (Patra et al., 2013; Lin et al., 2015) and for the Arctic (Bruhwiler et al., 2015), where seasonality (e.g. Zona et al., 2016, for tundra) and magnitude (e.g. Berchet et al., 2016, for continental shelves) of methane emissions remain poorly understood.

The ability to allocate observed atmospheric changes to changes of a given source is limited. Most inverse groups use EDGARv4.2 inventory as a prior, being the only annual gridded anthropogenic inventory to date. An updated version of the EDGARv4.2 inventory has been recently released (EDGARv4.2FT2012), which is very close at a global scale to the extrapolation performed in this paper based on statistics from BP (http://www.bp.com/) and on agriculture emissions from FAO (http://faostat3.fao.org). However, the significant changes in emissions in China (decrease) and Africa (increase) found in this synthesis strongly suggest the necessity to further revise the EDGAR inventory, in particular for coal-related emissions (China). Such an update is an ongoing effort in the EDGAR group. More extensive comparisons and exchange between the different inventory teams would also favour a path towards more consistency.

Uncertainties in the modelling of atmospheric transport and chemistry limit the optimal assimilation of atmospheric observations and increase the uncertainties of the inversionderived flux estimates. In this work, we gathered more inversion models than in Kirschke et al. (2013), leading to small to significant regional differences in the methane budget for 2000–2009. For the next release, it is important to stabilize the core group of participating inversions in order not to create artificial changes in the reporting of uncertainties. More, the recent results of Locatelli et al. (2015), who studied the sensitivity of inversion results to the representation of atmospheric transport, suggest that regional changes in the balance of methane emissions between inversions may be due to different characteristics of the transport models used here as compared to Kirschke et al. (2013). Indeed, the TRANSCOM experiment synthesized in Patra et al. (2011) showed a large sensitivity of the representation of atmospheric transport on methane concentrations in the atmosphere. As an illustration, in their study, the modelled CH₄ budget appeared to depend strongly on the troposphere– stratosphere exchange rate and thus on the model vertical grid structure and circulation in the lower stratosphere. These results put pressure to continue to improve atmospheric transport models, especially on the vertical.

We have built a global methane budget by gathering and synthesizing a large ensemble of published results using a consistent methodology, including atmospheric observations and inversions (top-down inversions), process-based models for land surface emissions and atmospheric chemistry, and inventories of anthropogenic emissions (bottom-up models and inventories). For the 2003–2012 decade, global methane emissions are 558 Tg CH₄ yr^-1 (range of 540–568), as estimated by top-down inversions. About 60 % of global emissions are anthropogenic (range of 50–70 %). Bottom-up models and inventories suggest much larger global emissions (736 Tg CH₄ yr^-1 [596–884]) mostly because of larger and more uncertain natural emissions from inland water systems, natural wetlands and geological leaks. Considering the atmospheric constraints on the top-down budget, it is likely that some of the individual emissions reported by the bottomup approaches are overestimated, leading to too large global emissions from a bottom-up perspective.

The latitudinal breakdown inferred from top-down approaches reveals a domination of tropical emissions (∼ 64 %) as compared to mid (∼ 32 %) and high (∼ 4 %) northern latitudes (above 60◦ N). The three largest emitting regions (South America, Africa, South East Asia) account for almost 50 % of the global budget. Top-down inversions consistently infer lower emissions in China (∼ 58 Tg CH₄ yr^-1 [51–72]) compared with the EDGARv4.2 inventories (> 70 Tg CH₄ yr^-1) but more consistent with the USEPA and GAINS inventories and with a recent regional inventory (∼ 45 Tg yr^-1). On the other hand, bottom-up methane emissions from Africa are lower than inferred from top-down inversions. These differences between top-down inversions and inventories call for a revisit of the emission factors and activity numbers used by the latter, at least for China and Africa.

Our results, including an extended set of inversions, are compared with the former synthesis of Kirschke et al. (2013), showing good consistency overall when comparing the same decade (2000–2009) at the global scale. Significant differences occur at the regional scale when comparing the 2000– 2009 decadal emissions. This important result indicates that using different transport models and inversion setups can significantly change the partition of emissions at the regional scale, making it less robust. It also means that we need to gather a stable, and as complete as possible, core of transport models in the next release of the budget in order to integrate this uncertainty within the budget.

Among the different uncertainties raised in Kirschke et al. (2013), the present work estimated that 30–40 % of the large range associated with modelled wetland emissions in Kirschke et al. (2013) was due to the estimation of wetland extent. The magnitudes and uncertainties of all other natural sources have been revised and updated, which has led to decreased the emission estimates for oceans, termites, wild animals and wildfires, and to increased emission estimates and range for freshwater systems. Although the risk of double counting emissions between natural and anthropogenic gas leaks exists, total fossil-related reported emissions are found consistent with atmospheric 14C. This places a clear priority on reducing uncertainties in emissions from inland water systems by better quantifying the emission factors of each contributor (streams, rivers, lakes, ponds) and eliminating the (plausible) double counting with wetland emissions. The development of process-based models for inland water emissions, constrained by observations, is a priority to overcome the present uncertainties on inland water emissions. Also important, although not addressed here, is to revise and update the magnitude, regional distribution, interannual variability and decadal trends in the OH radicals in the troposphere and stratosphere. This should be possible soon by the release of the CCMI ongoing multimodel intercomparison (http://www.igacproject.org/CCMI). Our work also suggests the need for more interactions among groups developing the emission inventories in order to resolve discrepancies on the magnitude of emissions and trends in key regions such as China or Africa. Particularly, the budget assessment of these regions should strongly benefit from the ongoing effort to develop a network of in situ atmospheric measurement stations. Finally, additional tracers (methane isotopes, ethane, CO) have potential to bring more constraint on the global methane cycle if their information content relative to methane emission trends is consistent with each other, which is not fully the case at present (Schaefer et al., 2016; Hausmann et al., 2016). Building on the improvement of the points above, our aim is to update this synthesis as a living review paper on a regular basis (∼ every 2 years). Each update will produce a more recent decadal CH₄ budget, highlight changes in emissions and trends, and show the availability and inclusion of new data, as well as model improvements.

On the top of the decadal methane budget presented in this paper, trends and year-to-year changes in the methane cycle have been highly discussed in the recent literature, especially because a sustained atmospheric positive growth rate of more than >+5 ppb yr^-1 has been observed since 2007 after almost a decade of stagnation in the late 1990s and early 2000s (Dlugokencky et al., 2011, Nisbet et al., 2014). Scenarios of increasing fossil and/or microbial sources have been proposed to explain this increase (Bousquet et al., 2011; Bergamaschi et al., 2013; Nisbet et al., 2014). Whereas the decreasing trend in δ¹3C in CH₄ suggests a significant, if not dominant, contribution from increasing emissions by microbial CH₄ sources (Schaefer et al., 2016; Nisbet et al., 2014), concurrent ethane and methane column measurements suggest a significant role (likely at least 39 %) for oil and gas production (Hausmann et al., 2016), which could be consistent when assuming a concomitant decrease in biomass burning emissions (heavy source for 13C), as suggested by the GFED database (Giglio et al., 2013). Yet accounting for the uncertainties in the isotopic signatures of the sources and their trends may suggest different portionings of the global methane sources between fossil fuel and biogenic methane emissions (Schwietzke et al., 2016). A possible positive OH trend has occurred since the 1970s followed by stagnation to decreasing OH in the 2000s, possibly contributing significantly to recent observed atmospheric methane changes (Dalsøren et al., 2016; Rigby et al., 2008; McNorton et al., 2016). The challenging increase of atmospheric methane during the past decade needs more efforts to be fully understood. GCP will take its part in analysing and synthesizing recent changes in the global to regional methane cycle based on the ensemble of top-down and bottom-up studies gathered for the budget analysis presented here.

The data presented here are made available in the belief that their wide dissemination will lead to greater understanding and new scientific insights on the methane budget and its changes and help to reduce the uncertainties in the methane budget. The free availability of the data does not constitute permission for publication of the data. For research projects, if the data used are essential to the work, or if the conclusion or results depend on the data, co-authorship may need to be considered. Full contact details and information on how to cite the data are given in the accompanying database.

The accompanying database includes one Excel file organized in the following spreadsheets and two netcdf files defining the regions used to produce the regional budget.

The file Global_Methane_Budget_2000-2012_v1.1.xlsx includes (1) a summary, (2) the methane observed mixing ratio and growth rate from the four global networks (NOAA, AGAGE, CSIRO and UCI), (3) the evolution of global anthropogenic methane emissions (excluding biomass burning emissions), used to produce Fig. 2, (4) the global and regional budgets over 2000–2009 based on bottom-up approaches, (5) the global and regional budgets over 2000– 2009 based on top-down approaches, (6) the global and regional budgets over 2003–2012 based on bottom-up approaches, (7) the global and regional budgets over 2003– 2012 based on top-down approaches, (8) the global and regional budgets for year 2012 based on bottom-up approaches, (9) the global and regional budgets for year 2012 based on top-down approaches, and (10) the list of contributors to contact for further information on specific data.

This database is available from the Carbon Dioxide Information Analysis Center (Saunois et al., 2016) and the Global Carbon Project (http://www.globalcarbonproject.org).

The Supplement related to this article is available at http://www.earth-syst-sci-data.net/8/697/2016/essd-8-697-2016-supplement.pdf